Interview – Customer Data Platform, more than CRM 2.0?

Interview with David M. Raab from the CDP Institute

David M. Raab is as a consultant specialized in marketing software and service vendor selection, marketing analytics and marketing technology assessment. Furthermore he is the founder of the Customer Data Platform Institute which is a vendor-neutral educational project to help marketers build a unified customer view that is available to all of their company systems.

Furthermore he is a Keynote-Speaker for the Predictive Analytics World Event 2019 in Berlin.

Data Science Blog: Mr. Raab, what exactly is a Customer Data Platform (CDP)? And where is the need for it?

The CDP Institute defines a Customer Data Platform as „packaged software that builds a unified, persistent customer database that is accessible by other systems“.  In plainer language, a CDP assembles customer data from all sources, combines it into customer profiles, and makes the profiles available for any use.  It’s important because customer data is collected in so many different systems today and must be unified to give customers the experience they expect.

Data Science Blog: Is it something like a CRM System 2.0? What Use Cases can be realized by a Customer Data Platform?

CRM systems are used to interact directly with customers, usually by telephone or in the field.  They work almost exclusively with data that is entered during those interactions.  This gives a very limited view of the customer since interactions through other channels such as order processing or Web sites are not included.  In fact, one common use case for CDP is to give CRM users a view of all customer interactions, typically by opening a window into the CDP database without needing to import the data into the CRM.  There are many other use cases for unified data, including customer segmentation, journey analysis, and personalization.  Anything that requires sharing data across different systems is a CDP use case.

Data Science Blog: When does a CDP make sense for a company? It is more relevant for retail and financial companies than for industrial companies, isn´t it?

CDP has been adopted most widely in retail and online media, where each customer has many interactions and there are many products to choose from.  This is a combination that can make good use of predictive modeling, which benefits greatly from having more complete data.  Financial services was slower to adopt, probably because they have fewer products but also because they already had pretty good customer data systems.  B2B has also been slow to adopt because so much of their customer relationship is handled by sales people.  We’ve more recently been seeing growth in additional sectors such as travel, healthcare, and education.  Those involve fewer transactions than retail but also rely on building strong customer relationships based on good data.

Data Science Blog: There are several providers for CDPs. Adobe, Tealium, Emarsys or Dynamic Yield, just to name some of them. Do they differ a lot between each other?

Yes they do.  All CDPs build the customer profiles I mentioned.  But some do more things, such as predictive modeling, message selection, and, increasingly, message delivery.  Of course they also vary in the industries they specialize in, regions they support, size of clients they work with, and many technical details.  This makes it hard to buy a CDP but also means buyers are more likely to find a system that fits their needs.

Data Science Blog: How established is the concept of the CDP in Europe in general? And how in comparison with the United States?

CDP is becoming more familiar in Europe but is not as well understood as in the U.S.  The European market spent a lot of money on Data Management Platforms (DMPs) which promised to do much of what a CDP does but were not able to because they do not store the level of detail that a CDP does.  Many DMPs also don’t work with personally identifiable data because the DMPs primarily support Web advertising, where many customers are anonymous.  The failures of DMPs have harmed CDPs because they have made buyers skeptical that any system can meet their needs, having already failed once.  But we are overcoming this as the market becomes better educated and more success stories are available.  What’s the same in Europe and the U.S. is that marketers face the same needs.  This will push European marketers towards CDPs as the best solution in many cases.

Data Science Blog: What are coming trends? What will be the main topic 2020?

We see many CDPs with broader functions for marketing execution: campaign management, personalization, and message delivery in particular.  This is because marketers would like to buy as few systems as possible, so they want broader scope in each systems.  We’re seeing expansion into new industries such as financial services, travel, telecommunications, healthcare, and education.  Perhaps most interesting will be the entry of Adobe, Salesforce, and Oracle, who have all promised CDP products late this year or early next year.  That will encourage many more people to consider buying CDPs.  We expect that market will expand quite rapidly, so current CDP vendors will be able to grow even as Adobe, Salesforce, and Oracle make new CDP sales.


You want to get in touch with Daniel M. Raab and understand more about the concept of a CDP? Meet him at the Predictive Analytics World 18th and 19th November 2019 in Berlin, Germany. As a Keynote-Speaker, he will introduce the concept of a Customer Data Platform in the light of Predictive Analytics. Click here to see the agenda of the event.

 


 

A Bird’s Eye View: How Machine Learning Can Help You Charge Your E-Scooters

Bird scooters in Columbus, Ohio

Bird scooters in Columbus, Ohio

Ever since I started using bike-sharing to get around in Seattle, I have become fascinated with geolocation data and the transportation sharing economy. When I saw this project leveraging the mobility data RESTful API from the Los Angeles Department of Transportation, I was eager to dive in and get my hands dirty building a data product utilizing a company’s mobility data API.

Unfortunately, the major bike and scooter providers (Bird, JUMP, Lime) don’t have publicly accessible APIs. However, some folks have seemingly been able to reverse-engineer the Bird API used to populate the maps in their Android and iOS applications.

One interesting feature of this data is the nest_id, which indicates if the Bird scooter is in a “nest” — a centralized drop-off spot for charged Birds to be released back into circulation.

I set out to ask the following questions:

  1. Can real-time predictions be made to determine if a scooter is currently in a nest?
  2. For non-nest scooters, can new nest location recommendations be generated from geospatial clustering?

To answer these questions, I built a full-stack machine learning web application, NestGenerator, which provides an automated recommendation engine for new nest locations. This application can help power Bird’s internal nest location generation that runs within their Android and iOS applications. NestGenerator also provides real-time strategic insight for Bird chargers who are enticed to optimize their scooter collection and drop-off route based on proximity to scooters and nest locations in their area.

Bird

The electric scooter market has seen substantial growth with Bird’s recent billion dollar valuation  and their $300 million Series C round in the summer of 2018. Bird offers electric scooters that top out at 15 mph, cost $1 to unlock and 15 cents per minute of use. Bird scooters are in over 100 cities globally and they announced in late 2018 that they eclipsed 10 million scooter rides since their launch in 2017.

Bird scooters in Tel Aviv, Israel

Bird scooters in Tel Aviv, Israel

With all of these scooters populating cities, there’s much-needed demand for people to charge them. Since they are electric, someone needs to charge them! A charger can earn additional income for charging the scooters at their home and releasing them back into circulation at nest locations. The base price for charging each Bird is $5.00. It goes up from there when the Birds are harder to capture.

Data Collection and Machine Learning Pipeline

The full data pipeline for building “NestGenerator”

Data

From the details here, I was able to write a Python script that returned a list of Bird scooters within a specified area, their geolocation, unique ID, battery level and a nest ID.

I collected scooter data from four cities (Atlanta, Austin, Santa Monica, and Washington D.C.) across varying times of day over the course of four weeks. Collecting data from different cities was critical to the goal of training a machine learning model that would generalize well across cities.

Once equipped with the scooter’s latitude and longitude coordinates, I was able to leverage additional APIs and municipal data sources to get granular geolocation data to create an original scooter attribute and city feature dataset.

Data Sources:

  • Walk Score API: returns a walk score, transit score and bike score for any location.
  • Google Elevation API: returns elevation data for all locations on the surface of the earth.
  • Google Places API: returns information about places. Places are defined within this API as establishments, geographic locations, or prominent points of interest.
  • Google Reverse Geocoding API: reverse geocoding is the process of converting geographic coordinates into a human-readable address.
  • Weather Company Data: returns the current weather conditions for a geolocation.
  • LocationIQ: Nearby Points of Interest (PoI) API returns specified PoIs or places around a given coordinate.
  • OSMnx: Python package that lets you download spatial geometries and model, project, visualize, and analyze street networks from OpenStreetMap’s APIs.

Feature Engineering

After extensive API wrangling, which included a four-week prolonged data collection phase, I was finally able to put together a diverse feature set to train machine learning models. I engineered 38 features to classify if a scooter is currently in a nest.

Full Feature Set

Full Feature Set

The features boiled down into four categories:

  • Amenity-based: parks within a given radius, gas stations within a given radius, walk score, bike score
  • City Network Structure: intersection count, average circuity, street length average, average streets per node, elevation level
  • Distance-based: proximity to closest highway, primary road, secondary road, residential road
  • Scooter-specific attributes: battery level, proximity to closest scooter, high battery level (> 90%) scooters within a given radius, total scooters within a given radius

 

Log-Scale Transformation

For each feature, I plotted the distribution to explore the data for feature engineering opportunities. For features with a right-skewed distribution, where the mean is typically greater than the median, I applied these log transformations to normalize the distribution and reduce the variability of outlier observations. This approach was used to generate a log feature for proximity to closest scooter, closest highway, primary road, secondary road, and residential road.

An example of a log transformation

Statistical Analysis: A Systematic Approach

Next, I wanted to ensure that the features I included in my model displayed significant differences when broken up by nest classification. My thinking was that any features that did not significantly differ when stratified by nest classification would not have a meaningful predictive impact on whether a scooter was in a nest or not.

Distributions of a feature stratified by their nest classification can be tested for statistically significant differences. I used an unpaired samples t-test with a 0.01% significance level to compute a p-value and confidence interval to determine if there was a statistically significant difference in means for a feature stratified by nest classification. I rejected the null hypothesis if a p-value was smaller than the 0.01% threshold and if the 99.9% confidence interval did not straddle zero. By rejecting the null-hypothesis in favor of the alternative hypothesis, it’s deemed there is a significant difference in means of a feature by nest classification.

Battery Level Distribution Stratified by Nest Classification to run a t-test

Battery Level Distribution Stratified by Nest Classification to run a t-test

Log of Closest Scooter Distribution Stratified by Nest Classification to run a t-test

Throwing Away Features

Using the approach above, I removed ten features that did not display statistically significant results.

Statistically Insignificant Features Removed Before Model Development

Model Development

I trained two models, a random forest classifier and an extreme gradient boosting classifier since tree-based models can handle skewed data, capture important feature interactions, and provide a feature importance calculation. I trained the models on 70% of the data collected for all four cities and reserved the remaining 30% for testing.

After hyper-parameter tuning the models for performance on cross-validation data it was time to run the models on the 30% of test data set aside from the initial data collection.

I also collected additional test data from other cities (Columbus, Fort Lauderdale, San Diego) not involved in training the models. I took this step to ensure the selection of a machine learning model that would generalize well across cities. The performance of each model on the additional test data determined which model would be integrated into the application development.

Performance on Additional Cities Test Data

The Random Forest Classifier displayed superior performance across the board

The Random Forest Classifier displayed superior performance across the board

I opted to move forward with the random forest model because of its superior performance on AUC score and accuracy metrics on the additional cities test data. AUC is the Area under the ROC Curve, and it provides an aggregate measure of model performance across all possible classification thresholds.

AUC Score on Test Data for each Model

AUC Score on Test Data for each Model

Feature Importance

Battery level dominated as the most important feature. Additional important model features were proximity to high level battery scooters, proximity to closest scooter, and average distance to high level battery scooters.

Feature Importance for the Random Forest Classifier

Feature Importance for the Random Forest Classifier

The Trade-off Space

Once I had a working machine learning model for nest classification, I started to build out the application using the Flask web framework written in Python. After spending a few days of writing code for the application and incorporating the trained random forest model, I had enough to test out the basic functionality. I could finally run the application locally to call the Bird API and classify scooter’s into nests in real-time! There was one huge problem, though. It took more than seven minutes to generate the predictions and populate in the application. That just wasn’t going to cut it.

The question remained: will this model deliver in a production grade environment with the goal of making real-time classifications? This is a key trade-off in production grade machine learning applications where on one end of the spectrum we’re optimizing for model performance and on the other end we’re optimizing for low latency application performance.

As I continued to test out the application’s performance, I still faced the challenge of relying on so many APIs for real-time feature generation. Due to rate-limiting constraints and daily request limits across so many external APIs, the current machine learning classifier was not feasible to incorporate into the final application.

Run-Time Compliant Application Model

After going back to the drawing board, I trained a random forest model that relied primarily on scooter-specific features which were generated directly from the Bird API.

Through a process called vectorization, I was able to transform the geolocation distance calculations utilizing NumPy arrays which enabled batch operations on the data without writing any “for” loops. The distance calculations were applied simultaneously on the entire array of geolocations instead of looping through each individual element. The vectorization implementation optimized real-time feature engineering for distance related calculations which improved the application response time by a factor of ten.

Feature Importance for the Run-time Compliant Random Forest Classifier

Feature Importance for the Run-time Compliant Random Forest Classifier

This random forest model generalized well on test-data with an AUC score of 0.95 and an accuracy rate of 91%. The model retained its prediction accuracy compared to the former feature-rich model, but it gained 60x in application performance. This was a necessary trade-off for building a functional application with real-time prediction capabilities.

Geospatial Clustering

Now that I finally had a working machine learning model for classifying nests in a production grade environment, I could generate new nest locations for the non-nest scooters. The goal was to generate geospatial clusters based on the number of non-nest scooters in a given location.

The k-means algorithm is likely the most common clustering algorithm. However, k-means is not an optimal solution for widespread geolocation data because it minimizes variance, not geodetic distance. This can create suboptimal clustering from distortion in distance calculations at latitudes far from the equator. With this in mind, I initially set out to use the DBSCAN algorithm which clusters spatial data based on two parameters: a minimum cluster size and a physical distance from each point. There were a few issues that prevented me from moving forward with the DBSCAN algorithm.

  1. The DBSCAN algorithm does not allow for specifying the number of clusters, which was problematic as the goal was to generate a number of clusters as a function of non-nest scooters.
  2. I was unable to hone in on an optimal physical distance parameter that would dynamically change based on the Bird API data. This led to suboptimal nest locations due to a distortion in how the physical distance point was used in clustering. For example, Santa Monica, where there are ~15,000 scooters, has a higher concentration of scooters in a given area whereas Brookline, MA has a sparser set of scooter locations.

An example of how sparse scooter locations vs. highly concentrated scooter locations for a given Bird API call can create cluster distortion based on a static physical distance parameter in the DBSCAN algorithm. Left:Bird scooters in Brookline, MA. Right:Bird scooters in Santa Monica, CA.

An example of how sparse scooter locations vs. highly concentrated scooter locations for a given Bird API call can create cluster distortion based on a static physical distance parameter in the DBSCAN algorithm. Left:Bird scooters in Brookline, MA. Right:Bird scooters in Santa Monica, CA.

Given the granularity of geolocation scooter data I was working with, geospatial distortion was not an issue and the k-means algorithm would work well for generating clusters. Additionally, the k-means algorithm parameters allowed for dynamically customizing the number of clusters based on the number of non-nest scooters in a given location.

Once clusters were formed with the k-means algorithm, I derived a centroid from all of the observations within a given cluster. In this case, the centroids are the mean latitude and mean longitude for the scooters within a given cluster. The centroids coordinates are then projected as the new nest recommendations.

NestGenerator showcasing non-nest scooters and new nest recommendations utilizing the K-Means algorithm

NestGenerator showcasing non-nest scooters and new nest recommendations utilizing the K-Means algorithm.

NestGenerator Application

After wrapping up the machine learning components, I shifted to building out the remaining functionality of the application. The final iteration of the application is deployed to Heroku’s cloud platform.

In the NestGenerator app, a user specifies a location of their choosing. This will then call the Bird API for scooters within that given location and generate all of the model features for predicting nest classification using the trained random forest model. This forms the foundation for map filtering based on nest classification. In the app, a user has the ability to filter the map based on nest classification.

Drop-Down Map View filtering based on Nest Classification

Drop-Down Map View filtering based on Nest Classification

Nearest Generated Nest

To see the generated nest recommendations, a user selects the “Current Non-Nest Scooters & Predicted Nest Locations” filter which will then populate the application with these nest locations. Based on the user’s specified search location, a table is provided with the proximity of the five closest nests and an address of the Nest location to help inform a Bird charger in their decision-making.

NestGenerator web-layout with nest addresses and proximity to nearest generated nests

NestGenerator web-layout with nest addresses and proximity to nearest generated nests

Conclusion

By accurately predicting nest classification and clustering non-nest scooters, NestGenerator provides an automated recommendation engine for new nest locations. For Bird, this application can help power their nest location generation that runs within their Android and iOS applications. NestGenerator also provides real-time strategic insight for Bird chargers who are enticed to optimize their scooter collection and drop-off route based on scooters and nest locations in their area.

Code

The code for this project can be found on my GitHub

Comments or Questions? Please email me an E-Mail!

 

Attribution Models in Marketing

Attribution Models

A Business and Statistical Case

INTRODUCTION

A desire to understand the causal effect of campaigns on KPIs

Advertising and marketing costs represent a huge and ever more growing part of the budget of companies. Studies have found out this share is as high as 10% and increases with the size of companies (CMO study by American Marketing Association and Duke University, 2017). Measuring precisely the impact of a specific marketing campaign on the sales of a company is a critical step towards an efficient allocation of this budget. Would the return be higher for an euro spent on a Facebook ad, or should we better spend it on a TV spot? How much should I spend on Twitter ads given the volume of sales this channel is responsible for?

Attribution Models have lately received great attention in Marketing departments to answer these issues. The transition from offline to online marketing methods has indeed permitted the collection of multiple individual data throughout the whole customer journey, and  allowed for the development of user-centric attribution models. In short, Attribution Models use the information provided by Tracking technologies such as Google Analytics or Webtrekk to understand customer journeys from the first click on a Facebook ad to the final purchase and adequately ponderate the different marketing campaigns encountered depending on their responsibility in the final conversion.

Issues on Causal Effects

A key question then becomes: how to declare a channel is responsible for a purchase? In other words, how can we isolate the causal effect or incremental value of a campaign ?

          1. A/B-Tests

One method to estimate the pure impact of a campaign is the design of randomized experiments, wherein a control and treated groups are compared.  A/B tests belong to this broad category of randomized methods. Provided the groups are a priori similar in every aspect except for the treatment received, all subsequent differences may be attributed solely to the treatment. This method is typically used in medical studies to assess the effect of a drug to cure a disease.

Main practical issues regarding Randomized Methods are:

  • Assuring that control and treated groups are really similar before treatment. Uually a random assignment (i.e assuring that on a relevant set of observable variables groups are similar) is realized;
  • Potential spillover-effects, i.e the possibility that the treatment has an impact on the non-treated group as well (Stable unit treatment Value Assumption, or SUTVA in Rubin’s framework);
  • The costs of conducting such an experiment, and especially the costs linked to the deliberate assignment of individuals to a group with potentially lower results;
  • The number of such experiments to design if multiple treatments have to be measured;
  • Difficulties taking into account the interaction effects between campaigns or the effect of spending levels. Indeed, usually A/B tests are led by cutting off temporarily one campaign entirely and measuring the subsequent impact on KPI’s compared to the situation where this campaign is maintained;
  • The dynamical reproduction of experiments if we assume that treatment effects may change over time.

In the marketing context, multiple campaigns must be tested in a dynamical way, and treatment effect is likely to be heterogeneous among customers, leading to practical issues in the lauching of A/B tests to approximate the incremental value of all campaigns. However, sites with a lot of traffic and conversions can highly benefit from A/B testing as it provides a scientific and straightforward way to approximate a causal impact. Leading companies such as Uber, Netflix or Airbnb rely on internal tools for A/B testing automation, which allow them to basically test any decision they are about to make.

References:

Books:

Experiment!: Website conversion rate optimization with A/B and multivariate testing, Colin McFarland, ©2013 | New Riders  

A/B testing: the most powerful way to turn clicks into customers. Dan Siroker, Pete Koomen; Wiley, 2013.

Blogs:

https://eng.uber.com/xp

https://medium.com/airbnb-engineering/growing-our-host-community-with-online-marketing-9b2302299324

Study:

https://cmosurvey.org/wp-content/uploads/sites/15/2018/08/The_CMO_Survey-Results_by_Firm_and_Industry_Characteristics-Aug-2018.pdf

        2. Attribution models

Attribution Models do not demand to create an experimental setting. They take into account existing data and derive insights from the variability of customer journeys. One key difficulty is then to differentiate correlation and causality in the links observed between the exposition to campaigns and purchases. Indeed, selection effects may bias results as exposure to campaigns is usually dependant on user-characteristics and thus may not be necessarily independant from the customer’s baseline conversion probabilities. For example, customers purchasing from a discount price comparison website may be intrinsically different from customers buying from FB ad and this a priori difference may alone explain post-exposure differences in purchasing bahaviours. This intrinsic weakness must be remembered when interpreting Attribution Models results.

                          2.1 General Issues

The main issues regarding the implementation of Attribution Models are linked to

  • Causality and fallacious reasonning, as most models do not take into account the aforementionned selection biases.
  • Their difficult evaluation. Indeed, in almost all attribution models (except for those based on classification, where the accuracy of the model can be computed), the additionnal value brought by the use of a given attribution models cannot be evaluated using existing historical data. This additionnal value can only be approximated by analysing how the implementation of the conclusions of the attribution model have impacted a given KPI.
  • Tracking issues, leading to an uncorrect reconstruction of customer journeys
    • Cross-device journeys: cross-device issue arises from the use of different devices throughout the customer journeys, making it difficult to link datapoints. For example, if a customer searches for a product on his computer but later orders it on his mobile, the AM would then mistakenly consider it an order without prior campaign exposure. Though difficult to measure perfectly, the proportion of cross-device orders can approximate 20-30%.
    • Cookies destruction makes it difficult to track the customer his the whole journey. Both regulations and consumers’ rising concerns about data privacy issues mitigate the reliability and use of cookies.1 – From 2002 on, the EU has enacted directives concerning privacy regulation and the extended use of cookies for commercial targeting purposes, which have highly impacted marketing strategies, such as the ‘Privacy and Electronic Communications Directive’ (2002/58/EC). A research was conducted and found out that the adoption of this ‘Privacy Directive’ had led to 64% decrease in advertising methods compared to the rest of the world (Goldfarb et Tucker (2011)). The effect was stronger for generalized sites (Yahoo) than for specialized sites.2 – Users have grown more and more conscious of data privacy issues and have adopted protective measures concerning data privacy, such as automatic destruction of cookies after a session is ended, or simply giving away less personnal information (Goldfarb et Tucker (2012) ) .Valuable user information may be lost, though tracking technologies evolution have permitted to maintain tracking by other means. This issue may be particularly important in countries highly concerned with data privacy issues such as Germany.
    • Offline/Online bridge: an Attribution Model should take into account all campaigns to draw valuable insights. However, the exposure to offline campaigns (TV, newspapers) are difficult to track at the user level. One idea to tackle this issue would be to estimate the proportion of conversions led by offline campaigns through AB testing and deduce this proportion from the credit assigned to the online campaigns accounted for in the Attribution Model.
    • Touch point information available: clicks are easy to follow but irrelevant to take into account the influence of purely visual campaigns such as display ads or video.

                          2.2 Today’s main practices

Two main families of Attribution Models exist:

  • Rule-Based Attribution Models, which have been used for in the last decade but from which companies are gradualy switching.

Attribution depends on the individual journeys that have led to a purchase and is solely based on the rank of the campaign in the journey. Some models focus on a single touch points (First Click, Last Click) while others account for multi-touch journeys (Bathtube, Linear). It can be calculated at the customer level and thus doesn’t require large amounts of data points. We can distinguish two sub-groups of rule-based Attribution Models:

  • One Touch Attribution Models attribute all credit to a single touch point. The First-Click model attributes all credit for a converion to the first touch point of the customer journey; last touch attributes all credit to the last campaign.
  • Multi-touch Rule-Based Attribution Models incorporate information on the whole customer journey are thus an improvement compared to one touch models. To this family belong Linear model where credit is split equally between all channels, Bathtube model where 40% of credit is given to first and last clicks and the remaining 20% is distributed equally between the middle channels, or time-decay models where credit assigned to a click diminishes as the time between the click and the order increases..

The main advantages of rule-based models is their simplicity and cost effectiveness. The main problems are:

– They are a priori known and can thus lead to optimization strategies from competitors
– They do not take into account aggregate intelligence on customer journeys and actual incremental values.
– They tend to bias (depending on the model chosen) channels that are over-represented at the beggining or end of the funnel, according to theoretical assumptions that have no observationnal back-ups.

  • Data-Driven Attribution Models

These models take into account the weaknesses of rule-based models and make a relevant use of available data. Being data-driven, following attribution models cannot be computed using single user level data. On the contrary values are calculated through data aggregation and thus require a certain volume of customer journey information.

References:

https://dspace.mit.edu/handle/1721.1/64920

 

        3. Data-Driven Attribution Models in practice

                          3.1 Issues

Several issues arise in the computation of campaigns individual impact on a given KPI within a data-driven model.

  • Selection biases: Exposure to certain types of advertisement is usually highly correlated to non-observable variables which are in turn correlated to consumption practices. Differences in the behaviour of users exposed to different campaigns may thus only be driven by core differences in conversion probabilities between groups whether than by the campaign effect.
  • Complementarity: it may be that campaigns A and B only have an effect when combined, so that measuring their individual impact would lead to misleading conclusions. The model could then try to assess the effect of combinations of campaigns on top of the effect of individual campaigns. As the number of possible non-ordered combinations of k campaigns is 2k, it becomes clear that inclusing all possible combinations would however be time-consuming.
  • Order-sensitivity: The effect of a campaign A may depend on the place where it appears in the customer journey, meaning the rank of a campaign and not merely its presence could be accounted for in the model.
  • Relative Order-sensitivity: it may be that campaigns A and B only have an effect when one is exposed to campaign A before campaign B. If so, it could be useful to assess the effect of given combinations of campaigns as well. And this for all campaigns, leading to tremendous numbers of possible combinations.
  • All previous phenomenon may be present, increasing even more the potential complexity of a comprehensive Attribution Model. The number of all possible ordered combination of k campaigns is indeed :

 

                          3.2 Main models

                                  A) Logistic Regression and Classification models

If non converting journeys are available, Attribition Model can be shaped as a simple classification issue. Campaign types or campaigns combination and volume of campaign types can be included in the model along with customer or time variables. As we are interested in inference (on campaigns effect) whether than prediction, a parametric model should be used, such as Logistic Regression. Non paramatric models such as Random Forests or Neural Networks can also be used though the interpretation of campaigns value would be more difficult to derive from the model results.

A common pitfall is the usual issue of spurious correlations on one hand and the correct interpretation of coefficients in business terms.

An advantage if the possibility to evaluate the relevance of the model using common model validation methods to evaluate its predictive power (validation set \ AUC \pseudo R squared).

                                  B) Shapley Value

Theory

The Shapley Value is based on a Game Theory framework and is named after its creator, the Nobel Price Laureate Lloyd Shapley. Initially meant to calculate the marginal contribution of players in cooperative games, the model has received much attention in research and industry and has lately been applied to marketing issues. This model is typically used by Google Adords and other ad bidding vendors. Campaigns or marketing channels are in this model seen as compementary players looking forward to increasing a given KPI.
Contrarily to Logistic Regressions, it is a non-parametric model. Contrarily to Markov Chains, all results are built using existing journeys, and not simulated ones.

Channels are considered to enter the game sequentially under a certain joining order. Shapley value try to The Shapley value of channel i is the weighted sum of the marginal values that channel i adds to all possible coalitions that don’t contain channel i.
In other words, the main logic is to analyse the difference of gains when a channel i is added after a coalition Ck of k channels, k<=n. We then sum all the marginal contributions over all possible ordered combination Ck of all campaigns excluding i, with k<=n-1.

Subsets framework

A first an most usual way to compute the Shapley Vaue is to consider that when a channel enters coalition, its additionnal value is the same irrelevant of the order in which previous channels have appeared. In other words, journeys (A>B>C) and (B>A>C) trigger the same gains.
Shapley value is computed as the gains associated to adding a channel i to a subset of channels, weighted by the number of (ordered) sequences that the (unordered) subset represents, summed up on all possible subsets of the total set of campaigns where the channel i is not present.
The Shapley value of the channel 𝑥𝑗 is then:

where |S| is the number of campaigns of a coalition S and the sum extends over all subsets S that do not not contain channel j. 𝜈(𝑆)  is the value of the coalition S and 𝜈(𝑆 ∪ {𝑥𝑗})  the value of the coalition formed by adding 𝑥𝑗 to coalition S. 𝜈(𝑆 ∪ {𝑥𝑗}) − 𝜈(𝑆) is thus the marginal contribution of channel 𝑥𝑗 to the coalition S.

The formula can be rewritten and understood as:

This method is convenient when data on the gains of on all possible permutations of all unordered k subsets of the n campaigns are available. It is also more convenient if the order of campaigns prior to the introduction of a campaign is thought to have no impact.

Ordered sequences

Let us define 𝜈((A>B)) as the value of the sequence A then B. What is we let 𝜈((A>B)) be different from 𝜈((B>A)) ?
This time we would need to sum over all possible permutation of the S campaigns present before  𝑥𝑗 and the N-(S+1) campaigns after 𝑥𝑗. Doing so we will sum over all possible orderings (i.e all permutations of the n campaigns of the grand coalition containing all campaigns) and we can remove the permutation coefficient s!(p-s+1)!.

This method is convenient when the order of channels prior to and after the introduction of another channel is assumed to have an impact. It is also necessary to possess data for all possible permutations of all k subsets of the n campaigns, and not only on all (unordered) k-subsets of the n campaigns, k<=n. In other words, one must know the gains of A, B, C, A>B, B>A, etc. to compute the Shapley Value.

Differences between the two approaches

We simulate an ordered case where the value for each ordered sequence k for k<=3 is known. We compare it to the usual Shapley value calculated based on known gains of unordered subsets of campaigns. So as to compare relevant values, we have built the gains matrix so that the gains of a subset A, B i.e  𝜈({B,A}) is the average of the gains of ordered sequences made up with A and B (assuming the number of journeys where A>B equals the number of journeys where B>A, we have 𝜈({B,A})=0.5( 𝜈((A>B)) + 𝜈((B>A)) ). We let the value of the grand coalition be different depending on the order of campaigns-keeping the constraints that it averages to the value used for the unordered case.

Note: mvA refers to the marginal value of A in a given sequence.
With traditionnal unordered coalitions:

With ordered sequences used to compute the marginal values:

 

We can see that the two approaches yield very different results. In the unordered case, the Shapley Value campaign C is the highest, culminating at 20, while A and B have the same Shapley Value mvA=mvB=15. In the ordered case, campaign A has the highest Shapley Value and all campaigns have different Shapley Values.

This example illustrates the inherent differences between the set and sequences approach to Shapley values. Real life data is more likely to resemble the ordered case as conversion probabilities may for any given set of campaigns be influenced by the order through which the campaigns appear.

Advantages

Shapley value has become popular in allocation problems in cooperative games because it is the unique allocation which satisfies different axioms:

  • Efficiency: Shaple Values of all channels add up to the total gains (here, orders) observed.
  • Symmetry: if channels A and B bring the same contribution to any coalition of campaigns, then their Shapley Value i sthe same
  • Null player: if a channel brings no additionnal gains to all coalitions, then its Shapley Value is zero
  • Strong monotony: the Shapley Value of a player increases weakly if all its marginal contributions increase weakly

These properties make the Shapley Value close to what we intuitively define as a fair attribution.

Issues

  • The Shapley Value is based on combinatory mathematics, and the number of possible coalitions and ordered sequences becomes huge when the number of campaigns increases.
  • If unordered, the Shapley Value assumes the contribution of campaign A is the same if followed by campaign B or by C.
  • If ordered, the number of combinations for which data must be available and sufficient is huge.
  • Channels rarely present or present in long journeys will be played down.
  • Generally, gains are supposed to grow with the number of players in the game. However, it is plausible that in the marketing context a journey with a high number of channels will not necessarily bring more orders than a journey with less channels involved.

References:

R package: GameTheoryAllocation

Article:
Zhao & al, 2018 “Shapley Value Methods for Attribution Modeling in Online Advertising “
https://link.springer.com/content/pdf/10.1007/s13278-017-0480-z.pdf
Courses: https://www.lamsade.dauphine.fr/~airiau/Teaching/CoopGames/2011/coopgames-7%5b8up%5d.pdf
Blogs: https://towardsdatascience.com/one-feature-attribution-method-to-supposedly-rule-them-all-shapley-values-f3e04534983d

                                  B) Markov Chains

Markov Chains are used to model random processes, i.e events that occur in a sequential manner and in such a way that the probability to move to a certain state only depends on the past steps. The number of previous steps that are taken into account to model the transition probability is called the memory parameter of the sequence, and for the model to have a solution must be comprised between 0 and 4. A Markov Chain process is thus defined entirely by its Transition Matrix and its initial vector (i.e the starting point of the process).

Markov Chains are applied in many scientific fields. Typically, they are used in weather forecasting, with the sequence of Sunny and Rainy days following a Markov Process of memory parameter 0, so that for each given day the probability that the next day will be rainy or sunny only depends on the weather of the current day. Other applications can be found in sociology to understand the dynamics of social classes intergenerational reproduction. To get more both mathematical and applied illustration, I recommend the reading of this course.

In the marketing context, Markov Chains are an interesting way to model the conversion funnel. To go from the from the Markov Model to the Attribution logic, we calculate the Removal Effect of each channel, i.e the difference in conversions that happen if the channel is removed. Please read below for an introduction to the methodology.

The first step in a Markov Chains Attribution Model is to build the transition matrix that captures the transition probabilities between the campaigns accross existing customer journeys. This Matrix is to be read as a “From state A to state B” table, from the left to the right. A first difficulty is finding the right memory parameter to use. A large memory parameter would allow to take more into account interraction effects within the conversion funnel but would lead to increased computationnal time, a non-readable transition matrix, and be more sensitive to noisy data. Please note that this transition matrix provides useful information on the conversion funnel and on the relationships between campaigns and can be used as such as an analytical tool. I suggest the clear and easily R code which can be found here or here.

Here is an illustration of a Markov Chain with memory Parameter of 0: the probability to go to a certain campaign B in the next step only depend on the campaign we are currently at:

The associated Transition Matrix is then (with null probabilities left as Blank):

The second step is  to compute the actual responsibility of a channel in total conversions. As mentionned above, the main philosophy to do so is to calculate the Removal Effect of each channel, i.e the changes in the number of conversions when a channel is entirely removed. All customer journeys which went through this channel are settled out to be unsuccessful. This calculation is done by applying the transition matrix with and without the removed channels to an initial vector that contains the number of desired simulations.

Building on our current example, we can then settle an initial vector with the desired number of simulations, e.g 10 000:

 

It is possible at this stage to add a constraint on the maximum number of times the matrix is applied to the data, i.e on the maximal number of campaigns a simulated journey is allowed to have.

Advantages

  • The dynamic journey is taken into account, as well as the transition between two states. The funnel is not assumed to be linear.
  • It is possile to build a conversion graph that maps the customer journey provides valuable insights.
  • It is possible to evaluate partly the accuracy of the Attribution Model based on Markov Chains. It is for example possible to see how well the transition matrix help predict the future by analysing the number of correct predictions at any given step over all sequences.

Disadvantages

  • It can be somewhat difficult to set the memory parameter. Complementarity effects between channels are not well taken into account if the memory is low, but a parameter too high will lead to over-sensitivity to noise in the data and be difficult to implement if customer journeys tend to have a number of campaigns below this memory parameter.
  • Long journeys with different channels involved will be overweighted, as they will count many times in the Removal Effect.  For example, if there are n-1 channels in the customer journey, this journey will be considered as failure for the n-1 channel-RE. If the volume effects (i.e the impact of the overall number of channels in a journey, irrelevant from their type° are important then results may be biased.

References:

R package: ChannelAttribution

Git:

https://github.com/MatCyt/Markov-Chain/blob/master/README.md

Course:

https://www.ssc.wisc.edu/~jmontgom/markovchains.pdf

Article:

“Mapping the Customer Journey: A Graph-Based Framework for Online Attribution Modeling”; Anderl, Eva and Becker, Ingo and Wangenheim, Florian V. and Schumann, Jan Hendrik, 2014. Available at SSRN: https://ssrn.com/abstract=2343077 or http://dx.doi.org/10.2139/ssrn.2343077

“Media Exposure through the Funnel: A Model of Multi-Stage Attribution”, Abhishek & al, 2012

“Multichannel Marketing Attribution Using Markov Chains”, Kakalejčík, L., Bucko, J., Resende, P.A.A. and Ferencova, M. Journal of Applied Management and Investments, Vol. 7 No. 1, pp. 49-60.  2018

Blogs:

https://analyzecore.com/2016/08/03/attribution-model-r-part-1

https://analyzecore.com/2016/08/03/attribution-model-r-part-2

                          3.3 To go further: Tackling selection biases with Quasi-Experiments

Exposure to certain types of advertisement is usually highly correlated to non-observable variables. Differences in the behaviour of users exposed to different campaigns may thus only be driven by core differences in converison probabilities between groups whether than by the campaign effect. These potential selection effects may bias the results obtained using historical data.

Quasi-Experiments can help correct this selection effect while still using available observationnal data.  These methods recreate the settings on a randomized setting. The goal is to come as close as possible to the ideal of comparing two populations that are identical in all respects except for the advertising exposure. However, populations might still differ with respect to some unobserved characteristics.

Common quasi-experimental methods used for instance in Public Policy Evaluation are:

  • Discontinuity Regressions
  • Matching Methods, such as Exact Matching,  Propensity-score matching or k-nearest neighbourghs.

References:

Article:

“Towards a digital Attribution Model: Measuring the impact of display advertising on online consumer behaviour”, Anindya Ghose & al, MIS Quarterly Vol. 40 No. 4, pp. 1-XX, 2016

https://pdfs.semanticscholar.org/4fa6/1c53f281fa63a9f0617fbd794d54911a2f84.pdf

        4. First Steps towards a Practical Implementation

Identify key points of interests

  • Identify the nature of touchpoints available: is the data based on clicks? If so, is there a way to complement the data with A/B tests to measure the influence of ads without clicks (display, video) ? For example, what happens to sales when display campaign is removed? Analysing this multiplier effect would give the overall responsibility of display on sales, to be deduced from current attribution values given to click-based channels. More interestingly, what is the impact of the removal of display campaign on the occurences of click-based campaigns ? This would give us an idea of the impact of display ads on the exposure to each other campaigns, which would help correct the attribution values more precisely at the campaign level.
  • Define the KPI to track. From a pure Marketing perspective, looking at purchases may be sufficient, but from a financial perspective looking at profits, though a bit more difficult to compute, may drive more interesting results.
  • Define a customer journey. It may seem obvious, but the notion needs to be clarified at first. Would it be defined by a time limit? If so, which one? Does it end when a conversion is observed? For example, if a customer makes 2 purchases, would the campaigns he’s been exposed to before the first order still be accounted for in the second order? If so, with a time decay?
  • Define the research framework: are we interested only in customer journeys which have led to conversions or in all journeys? Keep in mind that successful customer journeys are a non-representative sample of customer journeys. Models built on the analysis of biased samples may be conservative. Take an extreme example: 80% of customers who see campaign A buy the product, VS 1% for campaign B. However, campaign B exposure is great and 100 Million people see it VS only 1M for campaign A. An Attribution Model based on successful journeys will give higher credit to campaign B which is an auguable conclusion. Taking into account costs per campaign (in the case where costs are calculated by clicks) may of course tackle this issue partly, as campaign A could then exhibit higher returns, but a serious fallacious reasonning is at stake here.

Analyse the typical customer journey    

  • Performing a duration analysis on the data may help you improve the definition of the customer journey to be used by your organization. After which days are converison probabilities null? Should we consider the effect of campaigns disappears after x days without orders? For example, if 99% of orders are placed in the 30 days following a first click, it might be interesting to define the customer journey as a 30 days time frame following the first oder.
  • Look at the distribution of the number of campaigns in a typical journey. If you choose to calculate the effect of campaigns interraction in your Attribution Model, it may indeed help you determine the maximum number of campaigns to be included in a combination. Indeed, you may not need to assess the impact of channel combinations with above than 4 different channels if 95% of orders are placed after less then 4 campaigns.
  • Transition matrixes: what if a campaign A systematically leads to a campaign B? What happens if we remove A or B? These insights would give clues to ask precise questions for a latter AB test, for example to find out if there is complementarity between channels A and B – (implying none should be removed) or mere substitution (implying one can be given up).
  • If conversion rates are available: it can be interesting to perform a survival analysis i.e to analyse the likelihood of conversion based on duration since first click. This could help us excluse potential outliers or individuals who have very low conversion probabilities.

Summary

Attribution is a complex topic which will probably never be definitively solved. Indeed, a main issue is the difficulty, or even impossibility, to evaluate precisely the accuracy of the attribution model that we’ve built. Attribution Models should be seen as a good yet always improvable approximation of the incremental values of campaigns, and be presented with their intrinsinc limits and biases.

Introduction to ROC Curve

The abbreviation ROC stands for Receiver Operating Characteristic. Its main purpose is to illustrate the diagnostic ability of classifier as the discrimination threshold is varied. It was developed during World War II when Radar operators had to decide if the blip on the screen is an enemy target, a friendly ship or just a noise.  For these purposes they measured the ability of a radar receiver operator to make these important distinctions, which was called the Receiver Operating Characteristic.

Later it was found useful in interpreting medical test results and then in Machine learning classification problems. In order to get an introduction to binary classification and terms like ‘precision’ and ‘recall’ one can look into my earlier blog  here.

True positive rate and false positive rate

Let’s imagine a situation where a fire alarm is installed in a kitchen. The alarm is supposed to emit a sound in case fire smoke is detected in the room. Unfortunately, there is a lot of cooking done in the kitchen and the alarm may trigger the sound too often. Thus, instead of serving a purpose the alarm becomes a nuisance due to a large number of false alarms. In statistical terms these types of errors are called type 1 errors, or false positives.

One way to deal with this problem is to simply decrease sensitivity of the device. We do this by increasing the trigger threshold at the alarm setting. But then, not every alarm should have the same threshold setting. Consider the same type of device but kept in a bedroom. With high threshold, the device might miss smoke from a real short-circuit in the wires which poses a real danger of fire. This kind of failure is called Type 2 error or a false negative. Although the two devices are the same, different types of threshold settings are optimal for different circumstances.

To specify this more formally, let us describe the performance of a binary classifier at a particular threshold by the following parameters:

 

These parameters take different values at different thresholds. Hence, they define the performance of the classifier at particular threshold. But we want to examine in overall how good a classifier is. Fortunately, there is a way to do that. We plot the True Positive Rate (TPR) and False Positive rate (FPR) at different thresholds and this plot is called ROC curve.

Let’s try to understand this with an example.

A case with a distinct population distribution

Let’s suppose there is a disease which can be identified with deficiency of some parameter (maybe a certain vitamin). The distribution of population with this disease has a mean vitamin concentration sharply distinct from the mean of a healthy population, as shown below.

This is result of dummy data simulating population of 2000 people,the link to the code is given  in the end of this blog.  As the two populations are distinctly separated (there is no  overlap between the two distributions), we can expect that a classifier would have an easy job distinquishing healthy from sick people. We can run a logistic regression classifier with a threshold of .5 and be 100% succesful in detecting the decease.

The confusion matrix may look something like this.

In this ideal case with a threshold  of  .5 we do not make a single wrong classification. The True positive rate and False positive rate are 1 and 0, respectively. But we can shift the threshold. In that case, we will  get different confusion matrices. First we plot threshold vs. TPR.

We see for most values of threshold the TPR is close to 1 which again proves data is easy to classify and the classifier is returning high probabilities  for the most of positives .

Similarly Let’s plot threshold vs. FPR.

For most of the data points FPR is close to zero. This is also good. Now its time to plot the ROC curve using these results (TPR vs FPR).

Let’s try to interpret  the results,  all the points lie on line x=0 and y=1, it means for all the points FPR is zero or TPR is one, making  the curve a square. which means the classifier does perfectly well.

Case with overlapping  population distribution

The above example was about a perfect classifer. However, life is often not so easy. Now let us consider another more realistic situation in which the parameter distribution of the population is not as distinct as in the previous case. Rather, the mean of the parameter with healthy and not healthy datapoints are close and the distributions overlap, as shown in the next figure.

If we set the threshold to 0.5, the confusion matrix may look like this.

Now, any new choice of threshold location will affect both false positives and false negatives. In fact, there is a trade-off. If we shift the threshold with the goal to reduce false negatives, false positives will increase. If we move the threshold to the other direction and reduce false positive, false negatives will increase.

The plots (TPR vs Threshold) , (FPR vs Threshold) are shown below

If we plot the ROC curve from these results, it looks like this:

From the curve we see the classifier does not perform as well as the earlier one.

What else can be infered from this curve? We first need to understand what the diagonal in this plot represent. The diagonal represents ‘Line of no discrimination’, which we obtain if we randomly guess. This is the ROC curve for the worst possible classifier. Therefore, by comparing the obtained ROC curve with the diagonal, we see how much better our classifer is from random guessing.

The further away ROC curve from the diagonal is (the closest it is to the top left corner) , better the classifier is.

Area Under the curve

The overall performance of the classifier is given by the area under the ROC curve and is usually denoted as AUC. Since TPR and FPR lie within the range of 0 to 1, the AUC also assumes values between 0 and 1. The higher the value of AUC, the better is the overall performance of the classifier.

Let’s see this for the two different distributions which we saw earlier.

As we know the classifier had worked perfectly in the first case with points at (0,1) the area under the curve is 1 which is perfect. In the latter case the classifier was not able to perform as good, the ROC curve is between the diagonal and left hand corner. The AUC as we can see is less than 1.

Some other general characteristics

There are still few points that needs to be discussed on a General ROC curve

  • The ROC curve does not provide information about the actual values of thresholds used for the classifier.
  • Performance of different classifiers can be compared using the AUC of different Classifier. The larger the AUC, the better the classifier.
  • The vertical distance of the ROC curve from the no discrimination line gives a measure of ‘INFORMEDNESS’. This is known as Youden’s J satistic. This statistics can take values between 0 and 1.

Youden’s  J statistic is defined for every point on the ROC curve . The point at which Youden’s  J satistics reaches its maximum for a given ROC curve can be used to guide the selection of the threshold to be used for that classifier.

I hope this post does the job of providing an understanding of ROC curves  and AUC. The  Python program for simulating the example given earlier can be found here .

Please feel free to adjust the mean of the distributions and see the changes in the plot.

A common trap when it comes to sampling from a population that intrinsically includes outliers

I will discuss a common fallacy concerning the conclusions drawn from calculating a sample mean and a sample standard deviation and more importantly how to avoid it.

Suppose you draw a random sample x_1, x_2, … x_N of size N and compute the ordinary (arithmetic) sample mean  x_m and a sample standard deviation sd from it.  Now if (and only if) the (true) population mean µ (first moment) and population variance (second moment) obtained from the actual underlying PDF  are finite, the numbers x_m and sd make the usual sense otherwise they are misleading as will be shown by an example.

By the way: The common correlation coefficient will also be undefined (or in practice always point to zero) in the presence of infinite population variances. Hopefully I will create an article discussing this related fallacy in the near future where a suitable generalization to Lévy-stable variables will be proposed.

 Drawing a random sample from a heavy tailed distribution and discussing certain measures

As an example suppose you have a one dimensional random walker whose step length is distributed by a symmetric standard Cauchy distribution (Lorentz-profile) with heavy tails, i.e. an alpha-stable distribution with alpha being equal to one. The PDF of an individual independent step is given by p(x) = \frac{\pi^{-1}}{(1 + x^2)} , thus neither the first nor the second moment exist whereby the first exists and vanishes at least in the sense of a principal value due to symmetry.

Still let us generate N = 3000 (pseudo) standard Cauchy random numbers in R* to analyze the behavior of their sample mean and standard deviation sd as a function of the reduced sample size n \leq N.

*The R-code is shown at the end of the article.

Here are the piecewise sample mean (in blue) and standard deviation (in red) for the mentioned Cauchy sampling. We see that both the sample mean and sd include jumps and do not converge.

Especially the mean deviates relatively largely from zero even after 3000 observations. The sample sd has no target due to the population variance being infinite.

If the data is new and no prior distribution is known, computing the sample mean and sd will be misleading. Astonishingly enough the sample mean itself will have the (formally exact) same distribution as the single step length p(x). This means that the sample mean is also standard Cauchy distributed implying that with a different Cauchy sample one could have easily observed different sample means far of the presented values in blue.

What sense does it make to present the usual interval x_m \pm sd / \sqrt{N} in such a case? What to do?

The sample median, median absolute difference (mad) and Inter-Quantile-Range (IQR) are more appropriate to describe such a data set including outliers intrinsically. To make this plausible I present the following plot, whereby the median is shown in black, the mad in green and the IQR in orange.

This example shows that the median, mad and IQR converge quickly against their assumed values and contain no major jumps. These quantities do an obviously better job in describing the sample. Even in the presence of outliers they remain robust, whereby the mad converges more quickly than the IQR. Note that a standard Cauchy sample will contain half of its sample in the interval median \pm mad meaning that the IQR is twice the mad.

Drawing a random sample from a PDF that has finite moments

Just for comparison I also show the above quantities for a standard normal (pseudo) sample labeled with the same color as before as a counter example. In this case not only do both the sample mean and median but also the sd and mad converge towards their expected values (see plot below). Here all the quantities describe the data set properly and there is no trap since there are no intrinsic outliers. The sample mean itself follows a standard normal, so that the sd in deed makes sense and one could calculate a standard error \frac{sd}{\sqrt{N}} from it to present the usual stochastic confidence intervals for the sample mean.

A careful observation shows that in contrast to the Cauchy case here the sampled mean and sd converge more quickly than the sample median and the IQR. However still the sampled mad performs about as well as the sd. Again the mad is twice the IQR.

And here are the graphs of the prementioned quantities for a pseudo normal sample:

The take-home-message:

Just be careful when you observe outliers and calculate sample quantities right away, you might miss something. At best one carefully observes how the relevant quantities change with sample size as demonstrated in this article.

Such curves should become of broader interest in order to improve transparency in the Data Science process and reduce fallacies as well.

Thank you for reading.

P.S.: Feel free to play with the set random seed in the R-code below and observe how other quantities behave with rising sample size. Of course you can also try different PDFs at the beginning of the code. You can employ a Cauchy, Gaussian, uniform, exponential or Holtsmark (pseudo) random sample.

 

QUIZ: Which one of the recently mentioned random samples contains a trap** and why?

**in the context of this article

 

R-code used to generate the data and for producing plots:

 

#R-script for emphasizing convergence and divergence of sample means

####install and load relevant packages ####

#uncomment these lines if necessary
#install.packages(c('ggplot2',’stabledist’))
#library(ggplot2)
#library(stabledist)

#####drawing random samples #####

#Setting a random seed for being able to reproduce results  
set.seed(1234567)   
N= 2000     #sample size

#Choose a PDF from which a sample shall be drawn
#To do so (un)comment the respective lines of following code

data <- rcauchy(N)    # option1(default): standard Cauchy sampling

#data <- rnorm(N)     #option2: standard Gaussian sampling
                               
#data <- rexp(N)    # option3: standard exponential sampling

#data <- rstable(N,alpha=1.5,beta=0)  # option4: standard symmetric Holtsmark sampling

#data <- runif(N)              #option5: standard uniform sample

#####descriptive statistics####
#preparations/declarations

SUM = vector()
sd =vector()
mean = vector()
SQ =vector()
SQUARES = vector()
median = vector()
mad =vector()
quantiles = data.frame()
sem =vector()

#piecewise calculaion of descrptive quantities

for (k in 1:length(data)){              #mainloop
SUM[k] <- sum(data[1:k])            # sum of sample
mean[k] <- mean(data[1:k])          # arithmetic mean
sd[k] <- sd(data[1:k])              # standard deviation
sem[k] <- sd[k]/(sqrt(k))          #standard error of the sample mean (for finite variances)
mad[k] <- mad(data[1:k],const=1)   # median absolute deviation    

for (j in 1:5){
qq <- quantile(data[1:k],na.rm = T)
quantiles[k,j] <- qq[j]         #quantiles of sample
}
colnames(quantiles) <- c('min','Q1','median','Q3','max')

for (i in 1:length(data[1:k])){
SQUARES[i] <- data[i]*data[i]    
}
SQ[k] <- sum(SQUARES[1:k])    #sum of squares of random sample
}  #end of mainloop

#create table containing all relevant data
TABLE <-  as.data.frame(cbind(quantiles,mean,sd,SQ,SUM,sem))




#####plotting results###
x11()
print(ggplot(TABLE,aes(1:N,median))+
geom_point(size=.5)+xlab('sample size n')+ylab('sample median'))
x11()
print(ggplot(TABLE,aes(1:N,mad))+geom_point(size=.5,color ='green')+
xlab('sample size n')+ylab('sample median absolute difference'))
x11()
print(ggplot(TABLE,aes(1:N,sd))+geom_point(size=.5,color ='red')+
xlab('sample size n')+ylab('sample standard deviation'))
x11()
print(ggplot(TABLE,aes(1:N,mean))+geom_point(size=.5, color ='blue')+
xlab('sample size n')+ylab('sample mean'))
x11()
print(ggplot(TABLE,aes(1:N,Q3-Q1))+geom_point(size=.5, color ='blue')+
xlab('sample size n')+ylab('IQR'))

#uncomment the following lines of code to see further plots

#x11()
#print(ggplot(TABLE,aes(1:N,sem))+geom_point(size=.5)+
#xlab('sample size n')+ylab('sample sum of r.v.'))
#x11()
#print(ggplot(TABLE,aes(1:N,SUM))+geom_point(size=.5)+
#xlab('sample size n')+ylab('sample sum of r.v.'))
#x11()
#print(ggplot(TABLE,aes(1:N,SQ))+geom_point(size=.5)+
#xlab('sample size n')+ylab('sample sum of squares'))

 

Predictive maintenance in Semiconductor Industry: Part 1

The process in the semiconductor industry is highly complicated and is normally under consistent observation via the monitoring of the signals coming from several sensors. Thus, it is important for the organization to detect the fault in the sensor as quickly as possible. There are existing traditional statistical based techniques however modern semiconductor industries have the ability to produce more data which is beyond the capability of the traditional process.

For this article, we will be using SECOM dataset which is available here.  A lot of work has already done on this dataset by different authors and there are also some articles available online. In this article, we will focus on problem definition, data understanding, and data cleaning.

This article is only the first of three parts, in this article we will discuss the business problem in hand and clean the dataset. In second part we will do feature engineering and in the last article we will build some models and evaluate them.

Problem definition

This data which is collected by these sensors not only contains relevant information but also a lot of noise. The dataset contains readings from 590. Among the 1567 examples, there are only 104 fail cases which means that out target variable is imbalanced. We will look at the distribution of the dataset when we look at the python code.

NOTE: For a detailed description regarding this cases study I highly recommend to read the following research papers:

  •  Kerdprasop, K., & Kerdprasop, N. A Data Mining Approach to Automate Fault Detection Model Development in the Semiconductor Manufacturing Process.
  • Munirathinam, S., & Ramadoss, B. Predictive Models for Equipment Fault Detection in the Semiconductor Manufacturing Process.

Data Understanding and Preparation

Let’s start exploring the dataset now. The first step as always is to import the required libraries.

import pandas as pd
import numpy as np

There are several ways to import the dataset, you can always download and then import from your working directory. However, I will directly import using the link. There are two datasets: one contains the readings from the sensors and the other one contains our target variable and a timestamp.

# Load dataset
url = "https://archive.ics.uci.edu/ml/machine-learning-databases/secom/secom.data"
names = ["feature" + str(x) for x in range(1, 591)]
secom_var = pd.read_csv(url, sep=" ", names=names, na_values = "NaN") 


url_l = "https://archive.ics.uci.edu/ml/machine-learning-databases/secom/secom_labels.data"
secom_labels = pd.read_csv(url_l,sep=" ",names = ["classification","date"],parse_dates = ["date"],na_values = "NaN")

The first step before doing the analysis would be to merge the dataset and we will us pandas library to merge the datasets in just one line of code.

#Data cleaning
#1. Combined the two datasets
secom_merged = pd.merge(secom_var, secom_labels,left_index=True,right_index=True)

Now let’s check out the distribution of the target variable

secom_merged.target.value_counts().plot(kind = 'bar')

Figure 1: Distribution of Target Variable

From Figure 1 it can be observed that the target variable is imbalanced and it is highly recommended to deal with this problem before the model building phase to avoid bias model. Xgboost is one of the models which can deal with imbalance classes but one needs to spend a lot of time to tune the hyper-parameters to achieve the best from the model.

The dataset in hand contains a lot of null values and the next step would be to analyse these null values and remove the columns having null values more than a certain percentage. This percentage is calculated based on 95th quantile of null values.

#2. Analyzing nulls
secom_rmNa.isnull().sum().sum()
secom_nulls = secom_rmNa.isnull().sum()/len(secom_rmNa)
secom_nulls.describe()
secom_nulls.hist()

Figure 2: Missing percentge in each column

Now we calculate the 95th percentile of the null values.

x = secom_nulls.quantile(0.95)
secom_rmNa = secom_merged[secom_merged.columns[secom_nulls < x]]

Figure 3: Missing percentage after removing columns with more then 45% Na

From figure 3 its visible that there are still missing values in the dataset and can be dealt by using many imputation methods. The most common method is to impute these values by mean, median or mode. There also exist few sophisticated techniques like K-nearest neighbour and interpolation.  We will be applying interpolation technique to our dataset. 

secom_complete = secom_rmNa.interpolate()

To prepare our dataset for analysis we should remove some more unwanted columns like columns with near zero variance. For this we can calulate number of unique values in each column and if there is only one unique value we can delete the column as it holds no information.

df = secom_complete.loc[:,secom_complete.apply(pd.Series.nunique) != 1]

## Let's check the shape of the df
df.shape
(1567, 444)

We have applied few data cleaning techniques and reduced the features from 590 to 444. However, In the next article we will apply some feature engineering techniques and adress problems like the curse of dimensionality and will also try to balance the target variable.

Bleiben Sie dran!!

Fuzzy Matching mit dem Jaro-Winkler-Score zur Auswertung von Markenbekanntheit und Werbeerinnerung

Für Unternehmen sind Markenbekanntheit und Werbeerinnerung wichtige Zielgrößen, denn anhand dieser lässt sich ableiten, ob Konsumenten ein Produkt einer Marke kaufen werden oder nicht. Zielgrößen wie diese werden von Marktforschungsinstituten über Befragungen ermittelt. Dafür wird in regelmäßigen Zeitabständen eine gleichbleibende Anzahl an Personen befragt, ob diese sich an Marken einer bestimmten Branche erinnern oder sich an Werbung erinnern. Die Personen füllen dafür in der Regel einen Onlinefragebogen aus.

Die Ergebnisse der Befragung liegen in einer Datenmatrix (siehe Tabelle) vor und müssen zur Auswertung zunächst bearbeitet werden.

Laufende Nummer Marke 1 Marke 2 Marke 3 Marke 4
1 ING-Diba Citigroup Sparkasse
2 Sparkasse Consorsbank
3 Commerbank Deutsche Bank Sparkasse ING-DiBa
4 Sparkasse Targobank

Ziel ist es aus diesen Daten folgende 0/1 codierte Matrix zu generieren. Wenn eine Marke bekannt ist, wird in die zur Marke gehörende Spalte eine Eins eingetragen, ansonsten eine Null.

Alle Marken ING-Diba Citigroup Sparkasse Targobank
ING-Diba, Citigroup, Sparkasse 1 1 1 0
Sparkasse, Consorsbank 0 0 1 0
Commerzbank, Deutsche Bank, Sparkasse, ING-Diba 1 0 0 0
Sparkasse, Targobank 0 0 1 1

Der Workflow um diese Datentransformation durchzuführen ist oftmals mittels eines Teilstrings einer Marke zu suchen ob diese in einem über alle Nennungen hinweg zusammengeführten String vorkommt oder nicht (z.B. „argo“ bei Targobank). Das Problem dieser Herangehensweise ist, dass viele falsch geschriebenen Wörter so nicht erfasst werden und die Erfahrung zeigt, dass falsch geschriebene Marken in vielfältigster Weise auftreten. Hier mussten in der Vergangenheit Mitarbeiter sich in stundenlangem Kampf durch die Ergebnisse wühlen und falsch zugeordnete oder nicht zugeordnete Marken händisch korrigieren und alle Variationen der Wörter notieren, um für die nächste Befragung das Suchpattern zu optimieren.

Eine Alternative diesen aufwändigen Workflow stellt die Ermittlung von falsch geschriebenen Wörtern mittels des Jaro-Winkler-Scores dar. Dafür muss zunächst die Jaro-Winkler-Distanz zwischen zwei Strings berechnet werden. Diese berechnet sich wie folgt:

d_j = \frac{1}{3}(\frac{m}{|s_1|}+\frac{m}{|s_2|}+\frac{m - t}{m})

  • m: Anzahl der übereinstimmenden Buchstaben
  • s: Länge des Strings
  • t: Hälfte der Anzahl der Umstellungen der Buchstaben die nötig sind, damit Strings identisch sind. („Ta“ und „gobank“ befinden sich bereits in der korrekten Reihenfolge, somit gilt: t = 0)

Aus dem Ergebnis lässt sich der Jaro-Winkler Score berechnen:
d_w = \d_j + (l_p (1 - d_j))
ist dabei die Jaro-Winkler-Distanz, l die Länge der übereinstimmenden Buchstaben von Beginn des Wortes bis zum maximal vierten Buchstaben und p ein konstanter Faktor von 0,1.

Für die Strings „Targobank“ und „Tangobank“ ergibt sich die Jaro-Winkler-Distanz:

d_j = \frac{1}{3}(\frac{8}{9}+\frac{8}{9}+\frac{8 - 0}{9})

Daraus wird im nächsten Schritt der Jaro-Winkler Score berechnet:

d_w = 0,9259 + (2 \cdot 0,1 (1 - 0,9259)) = 0,9407407

Bisherige Erfahrungen haben gezeigt, dass sich Scores ab 0,8 bzw. 0,9 am besten zur Suche von ähnlichen Wörtern eignen. Ein Schwellenwert darunter findet sehr viele Wörter, die sich z.B. auch anderen Wörtern zuordnen lassen. Ein Schwellenwert über 0,9 identifiziert falsch geschriebene Wörter oftmals nicht mehr.

Nach diesem theoretischen Exkurs möchte ich nun zeigen, wie sich das Ganze praktisch anwenden lässt. Da sich das Ganze um ein fiktives Beispiel handelt, werden zur Demonstration der Praxistauglichkeit Fakedaten mit folgendem Code erzeugt. Dabei wird angenommen, dass Personen unterschiedlich viele Banken kennen und diese mit einer bestimmten Wahrscheinlichkeit falsch schreiben.

# Erstellung von Fakeantworten
set.seed(1234)
library(stringi)
library(tidyr)
library(RecordLinkage)
library(xlsx)
library(tm)
library(qdap)
library(stringr)
library(openxlsx)

konsonant <- c("r", "n", "g", "h", "b")
vokal <- c("a", "e", "o", "i", "u")

# Funktion, die mit einer zu bestimmenden Wahrscheinlichkeit, einen zufälligen Buchstaben erzeugt.
generate_wrong_words <- function(x, p, k = TRUE) {
  if(runif(1, 0, 1) > p) { # Zufallswert zwischen 0 und 1
    if(k == TRUE) { # Konsonant oder Vokal erzeugen
      string <- konsonant[sample.int(5, 1)] # Zufallszahl, die Index des Konsonnanten-Vektors bestimmt.
    } else {
      string <- vokal[sample.int(5, 1)] # Zufallszahl, die Index eines Vokal-Vecktors bestimmt.
    }
  } else {
    string <- x
  }
  return(string)
}

randombank <- function(x) {
  random_num <- runif(1, 0, 1)
  if(random_num  > x) { ## Wahrscheinlichkeit, dass Person keine Bank kennt.
    number <- sample.int(7, 1)
    if(number == 1) {
      bank <- paste0("Ta", generate_wrong_words(x = "r", p = 0.7), "gob", generate_wrong_words(x = "a", p = 0.9), "nk")
    } else if (number == 2) {
      bank <- paste0("Ing-di", generate_wrong_words(x = "b", p = 0.6), "a")
    } else if (number == 3) {
      bank <- paste0("com", generate_wrong_words(x = "m", p = 0.7), "erzb", generate_wrong_words(x = "a", p = 0.8), "nk")
    } else if (number == 4){
      bank <- paste0("Deutsch", generate_wrong_words(x = "e", p = 0.6, k = FALSE), " Ban", generate_wrong_words(x = "k", p = 0.8))
    } else if (number == 5) {
      bank <- paste0("Spark", generate_wrong_words(x = "a", p = 0.7, k = FALSE), "sse")
    } else if (number == 6) {
      bank <- paste0("Cons", generate_wrong_words(x = "o", p = 0.7, k = FALSE), "rsbank")
    } else {
      bank <- paste0("Cit", generate_wrong_words(x = "i", p = 0.7, k = FALSE), "gro", generate_wrong_words(x = "u", p = 0.9, k = FALSE), "p")
    }
  } else {
    bank <- "" # Leerer String, wenn keine Bank bekannt.
  }
  return(bank)
}


# DataFrame erzeugen, in dem Werte gespeichert werden.
df_raw <- data.frame(matrix(ncol = 8, nrow = 2500))

# Erzeugen von richtig und falsch geschrieben Banken mit einer durch bestimmten Variabilität an Banken, welche die Personen kennen.
for(i in 1:2500) {
  df_raw [i, 1] <- i # Laufende Nummer des Befragten
  df_raw [i, 2] <- randombank(x = 0.05)
  if(df_raw [i, 2] == "") { df_raw [i, 3] <- "" } else {df_raw [i, 3] <- randombank(x = 0.1)}
  if(df_raw [i, 3] == "") { df_raw [i, 4] <- "" } else {df_raw [i, 4] <- randombank(x = 0.1)}
  if(df_raw [i, 4] == "") { df_raw [i, 5] <- "" } else {df_raw [i, 5] <- randombank(x = 0.15)} 
  if(df_raw [i, 5] == "") { df_raw [i, 6] <- "" } else {df_raw [i, 6] <- randombank(x = 0.15)}
  if(df_raw [i, 6] == "") { df_raw [i, 7] <- "" } else {df_raw [i, 7] <- randombank(x = 0.2)} 
  if(df_raw [i, 7] == "") { df_raw [i, 8] <- "" } else {df_raw [i, 8] <- randombank(x = 0.2)} 
}
colnames(df_raw)[1] <- "lfdn"

Ausführen:

head(df_raw)

Nun werden die Inhalte der Spalten in eine einzige Spalte zusammengefasst und jede Marke per Komma getrennt.

df <- unite(df_raw, united, c(2:ncol(df_raw)), sep = ",")
colnames(df)[2] <- "text"
# Gesuchte Banken (nur korrekt geschrieben)
startliste <- c("Targobank", "Ing-DiBa", "Commerzbank", "Deutsche Bank", "Sparkasse", "Consorsbank", "Citigroup")

Damit Sonderzeichen, Leerzeichen oder Groß- und Kleinschreibung keine Rolle spielen, werden alle Strings vereinheitlicht und störende Zeichen entfernt.

dftext <- tolower(dftext)
dftext <- str_trim(dftext)
dftext <- gsub(" ", "", dftext)
dftext <- gsub("[?]", "", dftext)
dftext <- gsub("[-]", "", dftext)
dftext <- gsub("[_]", "", dftext)

startliste <- tolower(startliste)
startliste <- str_trim(startliste)
startliste <- gsub(" ", "", startliste)
startliste <- gsub("[?]", "", startliste)
startliste <- gsub("[-]", "", startliste)
startliste <- gsub("[_]", "", startliste)

Im nächsten Schritt wird geprüft welche Schreibweisen überhaupt existieren. Dafür eignet sich eine Word-Frequency-Matrix, mit der alle einzigartigen Wörter und deren Häufigkeiten in einem Vektor gezählt wird.

words <- as.data.frame(wfm(dftext)) # Jedes einzigartige Wort und dazugehörige Häufigkeiten. words <- rownames(words) # wfm zählt Häufigkeiten jedes Wortes und schreibt Wörter in rownames, wir brauchen jedoch das Wort selbst. </pre> Danach wird eine leere Liste erstellt, in der iterativ für jedes Element des Suchvektors ein Charactervektor erzeugt wird, der Wörter enthält, die einen Jaro-Winker Score von 0,9 oder höher besitzen. <pre class="theme:github lang:r decode:true ">for(i in 1:length(startliste)) {   finalewortliste[[i]] <- words[which(jarowinkler(startliste[[i]], words) > 0.9)] } </pre> Jetzt wird ein leerer DataFrame erzeugt, der die Zeilenlänge des originalen DataFrames besitzt sowie die Anzahl der Marken als Spaltenlänge. <pre class="theme:github lang:r decode:true ">finaldf <- data.frame(matrix(nrow = nrow(df), ncol = length(startliste))) colnames(finaldf) <- startliste </pre> Im nächsten Schritt wird nun aus den ähnlichen Wörtern mit einer oder-Verknüpfung einen String erzeugt, der alle durch den Jaro-Winkler-Score identifizierten Wörter beinhaltet. Wenn ein Treffer gefunden wird, wird in der Suchspalte eine Eins eingetragen, ansonsten eine Null. <pre class="theme:github lang:r decode:true ">for(i in 1:ncol(finaldf)) {   finaldf[i] <- ifelse(str_detect(dftext, paste(finalewortliste[[i]], collapse = "|")) == TRUE, 1, 0) 
}

Zuletzt wird eine Spalte erzeugt, in die eine Eins geschrieben wird, wenn keine der Marken gefunden wurde.

finaldfkeinedergeannten <- ifelse(rowSums(finaldf) > 0, 0, 1) # Wenn nicht mindestens eine der gesuchten Banken bekannt </pre> Nach der fertigen Berechnung der Matrix können nun die finalen KPI´s berechnet und als Report in eine .xlsx Datei geschrieben werden. <pre class="theme:github lang:r decode:true "># Prozentuale Anteile berechnen. anteil <- as.data.frame(t(sapply(finaldf, sum) / nrow(finaldf) * 100)) # Ordne dem DataFrame die ursprünglichen Nenneungen zu. finaldf <- cbind(dftext, finaldf)
colnames(finaldf)[1] <- "text"

# Ergebnisse in eine .xlsx Datei schreiben.
wb <- createWorkbook()
addWorksheet(wb, "Ergebnisse")    
writeData(wb, "Ergebnisse", anteil, startCol = 2, startRow = 1, rowNames = FALSE)
writeData(wb, "Ergebnisse", finaldf, startCol = 1, startRow = 4, rowNames = FALSE)
saveWorkbook(wb, paste0("C:/Users/User/Desktop/Results_", Sys.Date(), ".xlsx"), overwrite = TRUE)  

Dieses Vorgehen kann natürlich nicht verhindern, dass sich jemand mit kritischem Auge die Daten anschauen muss. In mehreren Tests ergaben sich bei einer Fallzahl von ~10.000 Antworten Genauigkeiten zwischen 95% und 100%, was bisherige Ansätze um ein Vielfaches übertrifft.9407407

Sentiment Analysis of IMDB reviews

Sentiment Analysis of IMDB reviews

This article shows you how to build a Neural Network from scratch(no libraries) for the purpose of detecting whether a movie review on IMDB is negative or positive.

Outline:

  • Curating a dataset and developing a "Predictive Theory"

  • Transforming Text to Numbers Creating the Input/Output Data

  • Building our Neural Network

  • Making Learning Faster by Reducing "Neural Noise"

  • Reducing Noise by strategically reducing the vocabulary

Curating the Dataset

In [3]:
def pretty_print_review_and_label(i):
    print(labels[i] + "\t:\t" + reviews[i][:80] + "...")

g = open('reviews.txt','r') # features of our dataset
reviews = list(map(lambda x:x[:-1],g.readlines()))
g.close()

g = open('labels.txt','r') # labels
labels = list(map(lambda x:x[:-1].upper(),g.readlines()))
g.close()

Note: The data in reviews.txt we're contains only lower case characters. That's so we treat different variations of the same word, like The, the, and THE, all the same way.

It's always a good idea to get check out your dataset before you proceed.

In [2]:
len(reviews) #No. of reviews
Out[2]:
25000
In [3]:
reviews[0] #first review
Out[3]:
'bromwell high is a cartoon comedy . it ran at the same time as some other programs about school life  such as  teachers  . my   years in the teaching profession lead me to believe that bromwell high  s satire is much closer to reality than is  teachers  . the scramble to survive financially  the insightful students who can see right through their pathetic teachers  pomp  the pettiness of the whole situation  all remind me of the schools i knew and their students . when i saw the episode in which a student repeatedly tried to burn down the school  i immediately recalled . . . . . . . . . at . . . . . . . . . . high . a classic line inspector i  m here to sack one of your teachers . student welcome to bromwell high . i expect that many adults of my age think that bromwell high is far fetched . what a pity that it isn  t   '
In [4]:
labels[0] #first label
Out[4]:
'POSITIVE'

Developing a Predictive Theory

Analysing how you would go about predicting whether its a positive or a negative review.

In [5]:
print("labels.txt \t : \t reviews.txt\n")
pretty_print_review_and_label(2137)
pretty_print_review_and_label(12816)
pretty_print_review_and_label(6267)
pretty_print_review_and_label(21934)
pretty_print_review_and_label(5297)
pretty_print_review_and_label(4998)
labels.txt 	 : 	 reviews.txt

NEGATIVE	:	this movie is terrible but it has some good effects .  ...
POSITIVE	:	adrian pasdar is excellent is this film . he makes a fascinating woman .  ...
NEGATIVE	:	comment this movie is impossible . is terrible  very improbable  bad interpretat...
POSITIVE	:	excellent episode movie ala pulp fiction .  days   suicides . it doesnt get more...
NEGATIVE	:	if you haven  t seen this  it  s terrible . it is pure trash . i saw this about ...
POSITIVE	:	this schiffer guy is a real genius  the movie is of excellent quality and both e...
In [41]:
from collections import Counter
import numpy as np

We'll create three Counter objects, one for words from postive reviews, one for words from negative reviews, and one for all the words.

In [56]:
# Create three Counter objects to store positive, negative and total counts
positive_counts = Counter()
negative_counts = Counter()
total_counts = Counter()

Examine all the reviews. For each word in a positive review, increase the count for that word in both your positive counter and the total words counter; likewise, for each word in a negative review, increase the count for that word in both your negative counter and the total words counter. You should use split(' ') to divide a piece of text (such as a review) into individual words.

In [57]:
# Loop over all the words in all the reviews and increment the counts in the appropriate counter objects
for i in range(len(reviews)):
    if(labels[i] == 'POSITIVE'):
        for word in reviews[i].split(" "):
            positive_counts[word] += 1
            total_counts[word] += 1
    else:
        for word in reviews[i].split(" "):
            negative_counts[word] += 1
            total_counts[word] += 1

Most common positive & negative words

In [ ]:
positive_counts.most_common()

The above statement retrieves alot of words, the top 3 being : ('the', 173324), ('.', 159654), ('and', 89722),

In [ ]:
negative_counts.most_common()

The above statement retrieves alot of words, the top 3 being : ('', 561462), ('.', 167538), ('the', 163389),

As you can see, common words like "the" appear very often in both positive and negative reviews. Instead of finding the most common words in positive or negative reviews, what you really want are the words found in positive reviews more often than in negative reviews, and vice versa. To accomplish this, you'll need to calculate the ratios of word usage between positive and negative reviews.

The positive-to-negative ratio for a given word can be calculated with positive_counts[word] / float(negative_counts[word]+1). Notice the +1 in the denominator – that ensures we don't divide by zero for words that are only seen in positive reviews.

In [58]:
pos_neg_ratios = Counter()

# Calculate the ratios of positive and negative uses of the most common words
# Consider words to be "common" if they've been used at least 100 times
for term,cnt in list(total_counts.most_common()):
    if(cnt > 100):
        pos_neg_ratio = positive_counts[term] / float(negative_counts[term]+1)
        pos_neg_ratios[term] = pos_neg_ratio

Examine the ratios

In [12]:
print("Pos-to-neg ratio for 'the' = {}".format(pos_neg_ratios["the"]))
print("Pos-to-neg ratio for 'amazing' = {}".format(pos_neg_ratios["amazing"]))
print("Pos-to-neg ratio for 'terrible' = {}".format(pos_neg_ratios["terrible"]))
Pos-to-neg ratio for 'the' = 1.0607993145235326
Pos-to-neg ratio for 'amazing' = 4.022813688212928
Pos-to-neg ratio for 'terrible' = 0.17744252873563218

We see the following:

  • Words that you would expect to see more often in positive reviews – like "amazing" – have a ratio greater than 1. The more skewed a word is toward postive, the farther from 1 its positive-to-negative ratio will be.
  • Words that you would expect to see more often in negative reviews – like "terrible" – have positive values that are less than 1. The more skewed a word is toward negative, the closer to zero its positive-to-negative ratio will be.
  • Neutral words, which don't really convey any sentiment because you would expect to see them in all sorts of reviews – like "the" – have values very close to 1. A perfectly neutral word – one that was used in exactly the same number of positive reviews as negative reviews – would be almost exactly 1.

Ok, the ratios tell us which words are used more often in postive or negative reviews, but the specific values we've calculated are a bit difficult to work with. A very positive word like "amazing" has a value above 4, whereas a very negative word like "terrible" has a value around 0.18. Those values aren't easy to compare for a couple of reasons:

  • Right now, 1 is considered neutral, but the absolute value of the postive-to-negative rations of very postive words is larger than the absolute value of the ratios for the very negative words. So there is no way to directly compare two numbers and see if one word conveys the same magnitude of positive sentiment as another word conveys negative sentiment. So we should center all the values around netural so the absolute value fro neutral of the postive-to-negative ratio for a word would indicate how much sentiment (positive or negative) that word conveys.
  • When comparing absolute values it's easier to do that around zero than one.

To fix these issues, we'll convert all of our ratios to new values using logarithms (i.e. use np.log(ratio))

In the end, extremely positive and extremely negative words will have positive-to-negative ratios with similar magnitudes but opposite signs.

In [59]:
# Convert ratios to logs
for word,ratio in pos_neg_ratios.most_common():
    pos_neg_ratios[word] = np.log(ratio)

Examine the new ratios

In [14]:
print("Pos-to-neg ratio for 'the' = {}".format(pos_neg_ratios["the"]))
print("Pos-to-neg ratio for 'amazing' = {}".format(pos_neg_ratios["amazing"]))
print("Pos-to-neg ratio for 'terrible' = {}".format(pos_neg_ratios["terrible"]))
Pos-to-neg ratio for 'the' = 0.05902269426102881
Pos-to-neg ratio for 'amazing' = 1.3919815802404802
Pos-to-neg ratio for 'terrible' = -1.7291085042663878

If everything worked, now you should see neutral words with values close to zero. In this case, "the" is near zero but slightly positive, so it was probably used in more positive reviews than negative reviews. But look at "amazing"'s ratio - it's above 1, showing it is clearly a word with positive sentiment. And "terrible" has a similar score, but in the opposite direction, so it's below -1. It's now clear that both of these words are associated with specific, opposing sentiments.

Run the below code to see more ratios.

It displays all the words, ordered by how associated they are with postive reviews.

In [ ]:
pos_neg_ratios.most_common()

The top most common words for the above code : ('edie', 4.6913478822291435), ('paulie', 4.0775374439057197), ('felix', 3.1527360223636558), ('polanski', 2.8233610476132043), ('matthau', 2.8067217286092401), ('victoria', 2.6810215287142909), ('mildred', 2.6026896854443837), ('gandhi', 2.5389738710582761), ('flawless', 2.451005098112319), ('superbly', 2.2600254785752498), ('perfection', 2.1594842493533721), ('astaire', 2.1400661634962708), ('captures', 2.0386195471595809), ('voight', 2.0301704926730531), ('wonderfully', 2.0218960560332353), ('powell', 1.9783454248084671), ('brosnan', 1.9547990964725592)

Transforming Text into Numbers

Creating the Input/Output Data

Create a set named vocab that contains every word in the vocabulary.

In [19]:
vocab = set(total_counts.keys())

Check vocabulary size

In [20]:
vocab_size = len(vocab)
print(vocab_size)
74074

Th following image rpresents the layers of the neural network you'll be building throughout this notebook. layer_0 is the input layer, layer_1 is a hidden layer, and layer_2 is the output layer.

In [1]:
 
Out[1]:

TODO: Create a numpy array called layer_0 and initialize it to all zeros. Create layer_0 as a 2-dimensional matrix with 1 row and vocab_size columns.

In [21]:
layer_0 = np.zeros((1,vocab_size))

layer_0 contains one entry for every word in the vocabulary, as shown in the above image. We need to make sure we know the index of each word, so run the following cell to create a lookup table that stores the index of every word.

TODO: Complete the implementation of update_input_layer. It should count how many times each word is used in the given review, and then store those counts at the appropriate indices inside layer_0.

In [ ]:
# Create a dictionary of words in the vocabulary mapped to index positions 
# (to be used in layer_0)
word2index = {}
for i,word in enumerate(vocab):
    word2index[word] = i

It stores the indexes like this: 'antony': 22, 'pinjar': 23, 'helsig': 24, 'dances': 25, 'good': 26, 'willard': 71500, 'faridany': 27, 'foment': 28, 'matts': 12313,

Lets implement some functions for simplifying our inputs to the neural network.

In [25]:
def update_input_layer(review):
    """
    The element at a given index of layer_0 should represent
    how many times the given word occurs in the review.
    """
     
    global layer_0
    
    # clear out previous state, reset the layer to be all 0s
    layer_0 *= 0
    
    # count how many times each word is used in the given review and store the results in layer_0 
    for word in review.split(" "):
        layer_0[0][word2index[word]] += 1

Run the following cell to test updating the input layer with the first review. The indices assigned may not be the same as in the solution, but hopefully you'll see some non-zero values in layer_0.

In [26]:
update_input_layer(reviews[0])
layer_0
Out[26]:
array([[ 18.,   0.,   0., ...,   0.,   0.,   0.]])

get_target_for_labels should return 0 or 1, depending on whether the given label is NEGATIVE or POSITIVE, respectively.

In [27]:
def get_target_for_label(label):
    if(label == 'POSITIVE'):
        return 1
    else:
        return 0

Building a Neural Network

In [32]:
import time
import sys
import numpy as np

# Encapsulate our neural network in a class
class SentimentNetwork:
    def __init__(self, reviews,labels,hidden_nodes = 10, learning_rate = 0.1):
        """
        Args:
            reviews(list) - List of reviews used for training
            labels(list) - List of POSITIVE/NEGATIVE labels
            hidden_nodes(int) - Number of nodes to create in the hidden layer
            learning_rate(float) - Learning rate to use while training
        
        """
        # Assign a seed to our random number generator to ensure we get
        # reproducable results
        np.random.seed(1)

        # process the reviews and their associated labels so that everything
        # is ready for training
        self.pre_process_data(reviews, labels)
        
        # Build the network to have the number of hidden nodes and the learning rate that
        # were passed into this initializer. Make the same number of input nodes as
        # there are vocabulary words and create a single output node.
        self.init_network(len(self.review_vocab),hidden_nodes, 1, learning_rate)

    def pre_process_data(self, reviews, labels):
        
        # populate review_vocab with all of the words in the given reviews
        review_vocab = set()
        for review in reviews:
            for word in review.split(" "):
                review_vocab.add(word)

        # Convert the vocabulary set to a list so we can access words via indices
        self.review_vocab = list(review_vocab)
        
        # populate label_vocab with all of the words in the given labels.
        label_vocab = set()
        for label in labels:
            label_vocab.add(label)
        
        # Convert the label vocabulary set to a list so we can access labels via indices
        self.label_vocab = list(label_vocab)
        
        # Store the sizes of the review and label vocabularies.
        self.review_vocab_size = len(self.review_vocab)
        self.label_vocab_size = len(self.label_vocab)
        
        # Create a dictionary of words in the vocabulary mapped to index positions
        self.word2index = {}
        for i, word in enumerate(self.review_vocab):
            self.word2index[word] = i
        
        # Create a dictionary of labels mapped to index positions
        self.label2index = {}
        for i, label in enumerate(self.label_vocab):
            self.label2index[label] = i
        
    def init_network(self, input_nodes, hidden_nodes, output_nodes, learning_rate):
        # Set number of nodes in input, hidden and output layers.
        self.input_nodes = input_nodes
        self.hidden_nodes = hidden_nodes
        self.output_nodes = output_nodes

        # Store the learning rate
        self.learning_rate = learning_rate

        # Initialize weights

        # These are the weights between the input layer and the hidden layer.
        self.weights_0_1 = np.zeros((self.input_nodes,self.hidden_nodes))
    
        # These are the weights between the hidden layer and the output layer.
        self.weights_1_2 = np.random.normal(0.0, self.output_nodes**-0.5, 
                                                (self.hidden_nodes, self.output_nodes))
        
        # The input layer, a two-dimensional matrix with shape 1 x input_nodes
        self.layer_0 = np.zeros((1,input_nodes))
    
    def update_input_layer(self,review):

        # clear out previous state, reset the layer to be all 0s
        self.layer_0 *= 0
        
        for word in review.split(" "):
            if(word in self.word2index.keys()):
                self.layer_0[0][self.word2index[word]] += 1
                
    def get_target_for_label(self,label):
        if(label == 'POSITIVE'):
            return 1
        else:
            return 0
        
    def sigmoid(self,x):
        return 1 / (1 + np.exp(-x))
    
    def sigmoid_output_2_derivative(self,output):
        return output * (1 - output)
    
    def train(self, training_reviews, training_labels):
        
        # make sure out we have a matching number of reviews and labels
        assert(len(training_reviews) == len(training_labels))
        
        # Keep track of correct predictions to display accuracy during training 
        correct_so_far = 0

        # Remember when we started for printing time statistics
        start = time.time()
        
        # loop through all the given reviews and run a forward and backward pass,
        # updating weights for every item
        for i in range(len(training_reviews)):
            
            # Get the next review and its correct label
            review = training_reviews[i]
            label = training_labels[i]
            
            ### Forward pass ###

            # Input Layer
            self.update_input_layer(review)

            # Hidden layer
            layer_1 = self.layer_0.dot(self.weights_0_1)

            # Output layer
            layer_2 = self.sigmoid(layer_1.dot(self.weights_1_2))
            
            ### Backward pass ###

            # Output error
            layer_2_error = layer_2 - self.get_target_for_label(label) # Output layer error is the difference between desired target and actual output.
            layer_2_delta = layer_2_error * self.sigmoid_output_2_derivative(layer_2)

            # Backpropagated error
            layer_1_error = layer_2_delta.dot(self.weights_1_2.T) # errors propagated to the hidden layer
            layer_1_delta = layer_1_error # hidden layer gradients - no nonlinearity so it's the same as the error

            # Update the weights
            self.weights_1_2 -= layer_1.T.dot(layer_2_delta) * self.learning_rate # update hidden-to-output weights with gradient descent step
            self.weights_0_1 -= self.layer_0.T.dot(layer_1_delta) * self.learning_rate # update input-to-hidden weights with gradient descent step

            # Keep track of correct predictions.
            if(layer_2 >= 0.5 and label == 'POSITIVE'):
                correct_so_far += 1
            elif(layer_2 < 0.5 and label == 'NEGATIVE'):
                correct_so_far += 1
            
            sys.stdout.write(" #Correct:" + str(correct_so_far) + " #Trained:" + str(i+1) \
                             + " Training Accuracy:" + str(correct_so_far * 100 / float(i+1))[:4] + "%")
    
    def test(self, testing_reviews, testing_labels):
        """
        Attempts to predict the labels for the given testing_reviews,
        and uses the test_labels to calculate the accuracy of those predictions.
        """
        
        # keep track of how many correct predictions we make
        correct = 0

        # Loop through each of the given reviews and call run to predict
        # its label. 
        for i in range(len(testing_reviews)):
            pred = self.run(testing_reviews[i])
            if(pred == testing_labels[i]):
                correct += 1
            
            sys.stdout.write(" #Correct:" + str(correct) + " #Tested:" + str(i+1) \
                             + " Testing Accuracy:" + str(correct * 100 / float(i+1))[:4] + "%")
    
    def run(self, review):
        """
        Returns a POSITIVE or NEGATIVE prediction for the given review.
        """
        # Run a forward pass through the network, like in the "train" function.
        
        # Input Layer
        self.update_input_layer(review.lower())

        # Hidden layer
        layer_1 = self.layer_0.dot(self.weights_0_1)

        # Output layer
        layer_2 = self.sigmoid(layer_1.dot(self.weights_1_2))
        
        # Return POSITIVE for values above greater-than-or-equal-to 0.5 in the output layer;
        # return NEGATIVE for other values
        if(layer_2[0] >= 0.5):
            return "POSITIVE"
        else:
            return "NEGATIVE"
        

Run the following code to create the network with a small learning rate, 0.001, and then train the new network. Using learning rate larger than this, for example 0.1 or even 0.01 would result in poor performance.

In [ ]:
mlp = SentimentNetwork(reviews[:-1000],labels[:-1000], learning_rate=0.001)
mlp.train(reviews[:-1000],labels[:-1000])

Running the above code would have given an accuracy around 62.2%

Reducing Noise in Our Input Data

Counting how many times each word occured in our review might not be the most efficient way. Instead just including whether a word was there or not will improve our training time and accuracy. Hence we update our update_input_layer() function.

In [ ]:
def update_input_layer(self,review):
    self.layer_0 *= 0
        
    for word in review.split(" "):
        if(word in self.word2index.keys()):
            self.layer_0[0][self.word2index[word]] =1

Creating and running our neural network again, even with a higher learning rate of 0.1 gave us a training accuracy of 83.8% and testing accuracy(testing on last 1000 reviews) of 85.7%.

Reducing Noise by Strategically Reducing the Vocabulary

Let us put the pos to neg ratio's that we found were much more effective at detecting a positive or negative label. We could do that by a few change:

  • Modify pre_process_data:
    • Add two additional parameters: min_count and polarity_cutoff
    • Calculate the positive-to-negative ratios of words used in the reviews.
    • Change so words are only added to the vocabulary if they occur in the vocabulary more than min_count times.
    • Change so words are only added to the vocabulary if the absolute value of their postive-to-negative ratio is at least polarity_cutoff
In [ ]:
def pre_process_data(self, reviews, labels, polarity_cutoff, min_count):
        
        positive_counts = Counter()
        negative_counts = Counter()
        total_counts = Counter()

        for i in range(len(reviews)):
            if(labels[i] == 'POSITIVE'):
                for word in reviews[i].split(" "):
                    positive_counts[word] += 1
                    total_counts[word] += 1
            else:
                for word in reviews[i].split(" "):
                    negative_counts[word] += 1
                    total_counts[word] += 1

        pos_neg_ratios = Counter()

        for term,cnt in list(total_counts.most_common()):
            if(cnt >= 50):
                pos_neg_ratio = positive_counts[term] / float(negative_counts[term]+1)
                pos_neg_ratios[term] = pos_neg_ratio

        for word,ratio in pos_neg_ratios.most_common():
            if(ratio > 1):
                pos_neg_ratios[word] = np.log(ratio)
            else:
                pos_neg_ratios[word] = -np.log((1 / (ratio + 0.01)))

        # populate review_vocab with all of the words in the given reviews
        review_vocab = set()
        for review in reviews:
            for word in review.split(" "):
                if(total_counts[word] > min_count):
                    if(word in pos_neg_ratios.keys()):
                        if((pos_neg_ratios[word] >= polarity_cutoff) or (pos_neg_ratios[word] <= -polarity_cutoff)):
                            review_vocab.add(word)
                    else:
                        review_vocab.add(word)

        # Convert the vocabulary set to a list so we can access words via indices
        self.review_vocab = list(review_vocab)
        
        # populate label_vocab with all of the words in the given labels.
        label_vocab = set()
        for label in labels:
            label_vocab.add(label)
        
        # Convert the label vocabulary set to a list so we can access labels via indices
        self.label_vocab = list(label_vocab)
        
        # Store the sizes of the review and label vocabularies.
        self.review_vocab_size = len(self.review_vocab)
        self.label_vocab_size = len(self.label_vocab)
        
        # Create a dictionary of words in the vocabulary mapped to index positions
        self.word2index = {}
        for i, word in enumerate(self.review_vocab):
            self.word2index[word] = i
        
        # Create a dictionary of labels mapped to index positions
        self.label2index = {}
        for i, label in enumerate(self.label_vocab):
            self.label2index[label] = i

Our training accuracy increased to 85.6% after this change. As we can see our accuracy saw a huge jump by making minor changes based on our intuition. We can keep making such changes and increase the accuracy even further.

 

Download the Data Sources

The data sources used in this article can be downloaded here:

The Inside Out of ML Based Prescriptive Analytics

With the constantly growing number of data, more and more companies are shifting towards analytic solutions. Analytic solutions help in extracting the meaning from the huge amount of data available. Thus, improving decision making.

Decision making is an important aspect of businesses, and technologies like Machine Learning are enhancing it further. The growing use of Machine Learning has changed the way of prescriptive analytics. In order to optimize the efforts, companies need to be more accurate with the historical and present data. This is because the historical and present data are the essentials of analytics. This article helps describe the inside out of Machine Learning-based prescriptive analytics.

Phases of business analytics

Descriptive analytics, predictive analytics, and prescriptive analytics are the three phases of business analytics. Descriptive analytics, being the first one, deals with past performance. Historical data is mined to understand past performance. This serves as a way to look for the reasons behind past success and failure. It is a kind of post-mortem analysis and most management reporting like sales, marketing, operations, and finance etc. make use of this.

The second one is a predictive analysis which answers the question of what is likely to happen. The historical data is now combined with rules, algorithms etc. to determine the possible future outcome or likelihood of a situation occurring.

The final phase, well known to everyone, is prescriptive analytics. It can continually take in new data and re-predict and re-prescribe. This improves the accuracy of the prediction and prescribes better decision options.  Professional services or technology or their combination can be chosen to perform all the three analytics.

More about prescriptive analytics

The analysis of business activities goes through many phases. Prescriptive analytics is one such. It is known to be the third phase of business analytics and comes after descriptive and predictive analytics. It entails the application of mathematical and computational sciences. It makes use of the results obtained from descriptive and predictive analysis to suggest decision options. It goes beyond predicting future outcomes and suggests actions to benefit from the predictions. It shows the implications of each decision option. It anticipates on what will happen when it will happen as well as why it will happen.

ML-based prescriptive analytics

Being just before the prescriptive analytics, predictive analytics is often confused with it. What actually happens is predictive analysis leads to prescriptive analysis. Thus, a Machine Learning based prescriptive analytics goes through an ML-based predictive analysis first. Therefore, it becomes necessary to consider the ML-based predictive analysis first.

ML-based predictive analytics:

A lot of things prevent businesses from achieving predictive analysis capabilities.  Machine Learning can be a great help in boosting Predictive analytics. Use of Machine Learning and Artificial Intelligence algorithms helps businesses in optimizing and uncovering the new statistical patterns. These statistical patterns form the backbone of predictive analysis. E-commerce, marketing, customer service, medical diagnosis etc. are some of the prospective use cases for Machine Learning based predictive analytics.

In E-commerce, machine learning can help in predicting the usual choices of the customer. Thus, presenting him/her according to his/her likes and dislikes. It can also help in predicting fraudulent transaction. Similarly, B2B marketing also makes good use of Machine learning based predictive analytics. Customer services and medical diagnosis also benefit from predictive analytics. Thus, a prediction and a prescription based on machine learning can boost various business functions.

Organizations and software development companies are making more and more use of machine learning based predictive analytics. The advancements like neural networks and deep learning algorithms are able to uncover hidden information. This all requires a well-researched approach. Big data and progressive IT systems also act as important factors in this.

Language Detecting with sklearn by determining Letter Frequencies

Of course, there are better and more efficient methods to detect the language of a given text than counting its lettes. On the other hand this is a interesting little example to show the impressing ability of todays machine learning algorithms to detect hidden patterns in a given set of data.

For example take the sentence:

“Ceci est une phrase française.”

It’s not to hard to figure out that this sentence is french. But the (lowercase) letters of the same sentence in a random order look like this:

“eeasrsçneticuaicfhenrpaes”

Still sure it’s french? Regarding the fact that this string contains the letter “ç” some people could have remembered long passed french lessons back in school and though might have guessed right. But beside the fact that the french letter “ç” is also present for example in portuguese, turkish, catalan and a few other languages, this is still a easy example just to explain the problem. Just try to guess which language might have generated this:

“ogldviisnntmeyoiiesettpetorotrcitglloeleiengehorntsnraviedeenltseaecithooheinsnstiofwtoienaoaeefiitaeeauobmeeetdmsflteightnttxipecnlgtetgteyhatncdisaceahrfomseehmsindrlttdthoaranthahdgasaebeaturoehtrnnanftxndaeeiposttmnhgttagtsheitistrrcudf”

While this looks simply confusing to the human eye and it seems practically impossible to determine the language it was generated from, this string still contains as set of hidden but well defined patterns from which the language could be predictet with almost complete (ca. 98-99%) certainty.

First of all, we need a set of texts in the languages our model should be able to recognise. Luckily with the package NLTK there comes a big set of example texts which actually are protocolls of the european parliament and therefor are publicly availible in 11 differen languages:

  •  Danish
  •  Dutch
  •  English
  •  Finnish
  •  French
  •  German
  •  Greek
  •  Italian
  •  Portuguese
  •  Spanish
  •  Swedish

Because the greek version is not written with the latin alphabet, the detection of the language greek would just be too simple, so we stay with the other 10 languages availible. To give you a idea of the used texts, here is a little sample:

“Resumption of the session I declare resumed the session of the European Parliament adjourned on Friday 17 December 1999, and I would like once again to wish you a happy new year in the hope that you enjoyed a pleasant festive period.
Although, as you will have seen, the dreaded ‘millennium bug’ failed to materialise, still the people in a number of countries suffered a series of natural disasters that truly were dreadful.”

Train and Test

The following code imports the nessesary modules and reads the sample texts from a set of text files into a pandas.Dataframe object and prints some statistics about the read texts:

from pathlib import Path
import random
from collections import Counter, defaultdict
import numpy as np
import pandas as pd
from sklearn.neighbors import *
from matplotlib import pyplot as plt
from mpl_toolkits import mplot3d
%matplotlib inline


def read(file):
    '''Returns contents of a file'''
    with open(file, 'r', errors='ignore') as f:
        text = f.read()
    return text

def load_eu_texts():
    '''Read texts snipplets in 10 different languages into pd.Dataframe

    load_eu_texts() -> pd.Dataframe
    
    The text snipplets are taken from the nltk-data corpus.
    '''
    basepath = Path('/home/my_username/nltk_data/corpora/europarl_raw/langs/')
    df = pd.DataFrame(columns=['text', 'lang', 'len'])
    languages = [None]
    for lang in basepath.iterdir():
        languages.append(lang.as_posix())
        t = '\n'.join([read(p) for p in lang.glob('*')])
        d = pd.DataFrame()
        d['text'] = ''
        d['text'] = pd.Series(t.split('\n'))
        d['lang'] = lang.name.title()
        df = df.append(d.copy(), ignore_index=True)
    return df

def clean_eutextdf(df):
    '''Preprocesses the texts by doing a set of cleaning steps
    
    clean_eutextdf(df) -> cleaned_df
    '''
    # Cuts of whitespaces a the beginning and and
    df['text'] = [i.strip() for i in df['text']]
    # Generate a lowercase Version of the text column
    df['ltext'] = [i.lower() for i in df['text']]

    # Determining the length of each text
    df['len'] = [len(i) for i in df['text']]
    # Drops all texts that are not at least 200 chars long
    df = df.loc[df['len'] > 200]
    return df

# Execute the above functions to load the texts
df = clean_eutextdf(load_eu_texts())

# Print a few stats of the read texts
textline = 'Number of text snippplets: ' + str(df.shape[0])
print('\n' + textline + '\n' + ''.join(['_' for i in range(len(textline))]))
c = Counter(df['lang'])
for l in c.most_common():
    print('%-25s' % l[0] + str(l[1]))
df.sample(10)
Number of text snippplets: 56481
________________________________
French                   6466
German                   6401
Italian                  6383
Portuguese               6147
Spanish                  6016
Finnish                  5597
Swedish                  4940
Danish                   4914
Dutch                    4826
English                  4791
lang	len	text	ltext
135233	Finnish	346	Vastustan sitä , toisin kuin tämän parlamentin...	vastustan sitä , toisin kuin tämän parlamentin...
170400	Danish	243	Desuden ødelægger det centraliserede europæisk...	desuden ødelægger det centraliserede europæisk...
85466	Italian	220	In primo luogo , gli accordi di Sharm el-Sheik...	in primo luogo , gli accordi di sharm el-sheik...
15926	French	389	Pour ce qui est concrètement du barrage de Ili...	pour ce qui est concrètement du barrage de ili...
195321	English	204	Discretionary powers for national supervisory ...	discretionary powers for national supervisory ...
160557	Danish	304	Det er de spørgmål , som de lande , der udgør ...	det er de spørgmål , som de lande , der udgør ...
196310	English	355	What remains of the concept of what a company ...	what remains of the concept of what a company ...
110163	Portuguese	327	Actualmente , é do conhecimento dos senhores d...	actualmente , é do conhecimento dos senhores d...
151681	Danish	203	Dette er vigtigt for den tillid , som samfunde...	dette er vigtigt for den tillid , som samfunde...
200540	English	257	Therefore , according to proponents , such as ...	therefore , according to proponents , such as ...

Above you see a sample set of random rows of the created Dataframe. After removing very short text snipplets (less than 200 chars) we are left with 56481 snipplets. The function clean_eutextdf() then creates a lower case representation of the texts in the coloum ‘ltext’ to facilitate counting the chars in the next step.
The following code snipplet now extracs the features – in this case the relative frequency of each letter in every text snipplet – that are used for prediction:

def calc_charratios(df):
    '''Calculating ratio of any (alphabetical) char in any text of df for each lyric
    
    calc_charratios(df) -> list, pd.Dataframe
    '''
    CHARS = ''.join({c for c in ''.join(df['ltext']) if c.isalpha()})
    print('Counting Chars:')
    for c in CHARS:
        print(c, end=' ')
        df[c] = [r.count(c) for r in df['ltext']] / df['len']
    return list(CHARS), df

features, df = calc_charratios(df)

Now that we have calculated the features for every text snipplet in our dataset, we can split our data set in a train and test set:

def split_dataset(df, ratio=0.5):
    '''Split the dataset into a train and a test dataset
    
    split_dataset(featuredf, ratio) -> pd.Dataframe, pd.Dataframe
    '''
    df = df.sample(frac=1).reset_index(drop=True)
    traindf = df[:][:int(df.shape[0] * ratio)]
    testdf = df[:][int(df.shape[0] * ratio):]
    return traindf, testdf

featuredf = pd.DataFrame()
featuredf['lang'] = df['lang']
for feature in features:
    featuredf[feature] = df[feature]
traindf, testdf = split_dataset(featuredf, ratio=0.80)

x = np.array([np.array(row[1:]) for index, row in traindf.iterrows()])
y = np.array([l for l in traindf['lang']])
X = np.array([np.array(row[1:]) for index, row in testdf.iterrows()])
Y = np.array([l for l in testdf['lang']])

After doing that, we can train a k-nearest-neigbours classifier and test it to get the percentage of correctly predicted languages in the test data set. Because we do not know what value for k may be the best choice, we just run the training and testing with different values for k in a for loop:

def train_knn(x, y, k):
    '''Returns the trained k nearest neighbors classifier
    
    train_knn(x, y, k) -> sklearn.neighbors.KNeighborsClassifier
    '''
    clf = KNeighborsClassifier(k)
    clf.fit(x, y)
    return clf

def test_knn(clf, X, Y):
    '''Tests a given classifier with a testset and return result
    
    text_knn(clf, X, Y) -> float
    '''
    predictions = clf.predict(X)
    ratio_correct = len([i for i in range(len(Y)) if Y[i] == predictions[i]]) / len(Y)
    return ratio_correct

print('''k\tPercentage of correctly predicted language
__________________________________________________''')
for i in range(1, 16):
    clf = train_knn(x, y, i)
    ratio_correct = test_knn(clf, X, Y)
    print(str(i) + '\t' + str(round(ratio_correct * 100, 3)) + '%')
k	Percentage of correctly predicted language
__________________________________________________
1	97.548%
2	97.38%
3	98.256%
4	98.132%
5	98.221%
6	98.203%
7	98.327%
8	98.247%
9	98.371%
10	98.345%
11	98.327%
12	98.3%
13	98.256%
14	98.274%
15	98.309%

As you can see in the output the reliability of the language classifier is generally very high: It starts at about 97.5% for k = 1, increases for with increasing values of k until it reaches a maximum level of about 98.5% at k ≈ 10.

Using the Classifier to predict languages of texts

Now that we have trained and tested the classifier we want to use it to predict the language of example texts. To do that we need two more functions, shown in the following piece of code. The first one extracts the nessesary features from the sample text and predict_lang() predicts the language of a the texts:

def extract_features(text, features):
    '''Extracts all alphabetic characters and add their ratios as feature
    
    extract_features(text, features) -> np.array
    '''
    textlen = len(text)
    ratios = []
    text = text.lower()
    for feature in features:
        ratios.append(text.count(feature) / textlen)
    return np.array(ratios)

def predict_lang(text, clf=clf):
    '''Predicts the language of a given text and classifier
    
    predict_lang(text, clf) -> str
    '''
    extracted_features = extract_features(text, features)
    return clf.predict(np.array(np.array([extracted_features])))[0]

text_sample = df.sample(10)['text']

for example_text in text_sample:
    print('%-20s'  % predict_lang(example_text, clf) + '\t' + example_text[:60] + '...')
Italian             	Auspico che i progetti riguardanti i programmi possano contr...
English             	When that time comes , when we have used up all our resource...
Portuguese          	Creio que o Parlamento protesta muitas vezes contra este mét...
Spanish             	Sobre la base de esta posición , me parece que se puede enco...
Dutch               	Ik voel mij daardoor aangemoedigd omdat ik een brede consens...
Spanish             	Señor Presidente , Señorías , antes que nada , quisiera pron...
Italian             	Ricordo altresì , signora Presidente , che durante la preced...
Swedish             	Betänkande ( A5-0107 / 1999 ) av Berend för utskottet för re...
English             	This responsibility cannot only be borne by the Commissioner...
Portuguese          	A nossa leitura comum é que esse partido tem uma posição man...

With this classifier it is now also possible to predict the language of the randomized example snipplet from the introduction (which is acutally created from the first paragraph of this article):

example_text = "ogldviisnntmeyoiiesettpetorotrcitglloeleiengehorntsnraviedeenltseaecithooheinsnstiofwtoienaoaeefiitaeeauobmeeetdmsflteightnttxipecnlgtetgteyhatncdisaceahrfomseehmsindrlttdthoaranthahdgasaebeaturoehtrnnanftxndaeeiposttmnhgttagtsheitistrrcudf"
predict_lang(example_text)
'English'

The KNN classifier of sklearn also offers the possibility to predict the propability with which a given classification is made. While the probability distribution for a specific language is relativly clear for long sample texts it decreases noticeably the shorter the texts are.

def dict_invert(dictionary):
    ''' Inverts keys and values of a dictionary
    
    dict_invert(dictionary) -> collections.defaultdict(list)
    '''
    inverse_dict = defaultdict(list)
    for key, value in dictionary.items():
        inverse_dict[value].append(key)
    return inverse_dict

def get_propabilities(text, features=features):
    '''Prints the probability for every language of a given text
    
    get_propabilities(text, features)
    '''
    results = clf.predict_proba(extract_features(text, features=features).reshape(1, -1))
    for result in zip(clf.classes_, results[0]):
        print('%-20s' % result[0] + '%7s %%' % str(round(float(100 * result[1]), 4)))


example_text = 'ogldviisnntmeyoiiesettpetorotrcitglloeleiengehorntsnraviedeenltseaecithooheinsnstiofwtoienaoaeefiitaeeauobmeeetdmsflteightnttxipecnlgtetgteyhatncdisaceahrfomseehmsindrlttdthoaranthahdgasaebeaturoehtrnnanftxndaeeiposttmnhgttagtsheitistrrcudf'
print(example_text)
get_propabilities(example_text + '\n')
print('\n')
example_text2 = 'Dies ist ein kurzer Beispielsatz.'
print(example_text2)
get_propabilities(example_text2 + '\n')
ogldviisnntmeyoiiesettpetorotrcitglloeleiengehorntsnraviedeenltseaecithooheinsnstiofwtoienaoaeefiitaeeauobmeeetdmsflteightnttxipecnlgtetgteyhatncdisaceahrfomseehmsindrlttdthoaranthahdgasaebeaturoehtrnnanftxndaeeiposttmnhgttagtsheitistrrcudf
Danish                  0.0 %
Dutch                   0.0 %
English               100.0 %
Finnish                 0.0 %
French                  0.0 %
German                  0.0 %
Italian                 0.0 %
Portuguese              0.0 %
Spanish                 0.0 %
Swedish                 0.0 %


Dies ist ein kurzer Beispielsatz.
Danish                  0.0 %
Dutch                   0.0 %
English                 0.0 %
Finnish                 0.0 %
French              18.1818 %
German              72.7273 %
Italian              9.0909 %
Portuguese              0.0 %
Spanish                 0.0 %
Swedish                 0.0 %

Background and Insights

Why does a relative simple model like counting letters acutally work? Every language has a specific pattern of letter frequencies which can be used as a kind of fingerprint: While there are almost no y‘s in the german language this letter is quite common in english. In french the letter k is not very common because it is replaced with q in most cases.

For a better understanding look at the output of the following code snipplet where only three letters already lead to a noticable form of clustering:

projection='3d')
legend = []
X, Y, Z = 'e', 'g', 'h'

def iterlog(ln):
    retvals = []
    for n in ln:
        try:
            retvals.append(np.log(n))
        except:
            retvals.append(None)
    return retvals

for X in ['t']:
    ax = plt.axes(projection='3d')
    ax.xy_viewLim.intervalx = [-3.5, -2]
    legend = []
    for lang in [l for l in df.groupby('lang') if l[0] in {'German', 'English', 'Finnish', 'French', 'Danish'}]:
        sample = lang[1].sample(4000)

        legend.append(lang[0])
        ax.scatter3D(iterlog(sample[X]), iterlog(sample[Y]), iterlog(sample[Z]))

    ax.set_title('log(10) of the Relativ Frequencies of "' + X.upper() + "', '" + Y.upper() + '" and "' + Z.upper() + '"\n\n')
    ax.set_xlabel(X.upper())
    ax.set_ylabel(Y.upper())
    ax.set_zlabel(Z.upper())
    plt.legend(legend)
    plt.show()

 

Even though every single letter frequency by itself is not a very reliable indicator, the set of frequencies of all present letters in a text is a quite good evidence because it will more or less represent the letter frequency fingerprint of the given language. Since it is quite hard to imagine or visualize the above plot in more than three dimensions, I used a little trick which shows that every language has its own typical fingerprint of letter frequencies:

legend = []
fig = plt.figure(figsize=(15, 10))
plt.axes(yscale='log')
    
langs = defaultdict(list)

for lang in [l for l in df.groupby('lang') if l[0] in set(df['lang'])]:
    for feature in 'abcdefghijklmnopqrstuvwxyz':
        langs[lang[0]].append(lang[1][feature].mean())

mean_frequencies = {feature:df[feature].mean() for feature in 'abcdefghijklmnopqrstuvwxyz'}
for i in langs.items():
    legend.append(i[0])
    j = np.array(i[1]) / np.array([mean_frequencies[c] for c in 'abcdefghijklmnopqrstuvwxyz'])
    plt.plot([c for c in 'abcdefghijklmnopqrstuvwxyz'], j)
plt.title('Log. of relative Frequencies compared to the mean Frequency in all texts')
plt.xlabel('Letters')
plt.ylabel('(log(Lang. Frequencies / Mean Frequency)')
plt.legend(legend)
plt.grid()
plt.show()

What more?

Beside the fact, that letter frequencies alone, allow us to predict the language of every example text (at least in the 10 languages with latin alphabet we trained for) with almost complete certancy there is even more information hidden in the set of sample texts.

As you might know, most languages in europe belong to either the romanian or the indogermanic language family (which is actually because the romans conquered only half of europe). The border between them could be located in belgium, between france and germany and in swiss. West of this border the romanian languages, which originate from latin, are still spoken, like spanish, portouguese and french. In the middle and northern part of europe the indogermanic languages are very common like german, dutch, swedish ect. If we plot the analysed languages with a different colour sheme this border gets quite clear and allows us to take a look back in history that tells us where our languages originate from:

legend = []
fig = plt.figure(figsize=(15, 10))
plt.axes(yscale='linear')
    
langs = defaultdict(list)
for lang in [l for l in df.groupby('lang') if l[0] in {'German', 'English', 'French', 'Spanish', 'Portuguese', 'Dutch', 'Swedish', 'Danish', 'Italian'}]:
    for feature in 'abcdefghijklmnopqrstuvwxyz':
        langs[lang[0]].append(lang[1][feature].mean())

colordict = {l[0]:l[1] for l in zip([lang for lang in langs], ['brown', 'tomato', 'orangered',
                                                               'green', 'red', 'forestgreen', 'limegreen',
                                                               'darkgreen', 'darkred'])}
mean_frequencies = {feature:df[feature].mean() for feature in 'abcdefghijklmnopqrstuvwxyz'}
for i in langs.items():
    legend.append(i[0])
    j = np.array(i[1]) / np.array([mean_frequencies[c] for c in 'abcdefghijklmnopqrstuvwxyz'])
    plt.plot([c for c in 'abcdefghijklmnopqrstuvwxyz'], j, color=colordict[i[0]])
#     plt.plot([c for c in 'abcdefghijklmnopqrstuvwxyz'], i[1], color=colordict[i[0]])
plt.title('Log. of relative Frequencies compared to the mean Frequency in all texts')
plt.xlabel('Letters')
plt.ylabel('(log(Lang. Frequencies / Mean Frequency)')
plt.legend(legend)
plt.grid()
plt.show()

As you can see the more common letters, especially the vocals like a, e, i, o and u have almost the same frequency in all of this languages. Far more interesting are letters like q, k, c and w: While k is quite common in all of the indogermanic languages it is quite rare in romanic languages because the same sound is written with the letters q or c.
As a result it could be said, that even “boring” sets of data (just give it a try and read all the texts of the protocolls of the EU parliament…) could contain quite interesting patterns which – in this case – allows us to predict quite precisely which language a given text sample is written in, without the need of any translation program or to speak the languages. And as an interesting side effect, where certain things in history happend (or not happend): After two thousand years have passed, modern machine learning techniques could easily uncover this history because even though all these different languages developed, they still have a set of hidden but common patterns that since than stayed the same.