Visual Feature Engineering for Machine Learning with React

Tensorflow.js brought Machine Learning to Javascript that gives a wide range of possibilities. Now it is possible to build, train and use ML models directly in browser. But the efficiency of ML model highly depends on the features that it is using. React allows to reinforce ML feature engineering process. It is great for displaying fast-changing data in web and let to visualize features, model’s layers and weights in a run-time during model training that gives useful insides on how features correspond to the result, what are good features and what features could be removed. During the session we will use React-based model visualization to pick the best features to predict results of soccer matches played at the FIFA World Cups. We will check count of matches that team have won and lost, count of wins and loses against a particular competitor, goal differences, stage, etc. We will validate our model using matches results from the last Mundial in Russia in 2018.