Quick Start: follow this link.
Getting Started: follow this tutorial.
Titanic Problem: From automatic feature engineering to automatic feature selection, but only for a given test dataset. (Note that this work is based on Featuretools version 0.27.1)
Titanic Automation: Model Generation with Training Data
- FeatureTools automatized FE, saved FE
- Convert categorical data to numbers, saved in pickle
- Feature selecion (Remove collinear features), saved in pickle
- Modeling (random forest), saved in pickle
Model Generation version II: feature selection using Recursive Feature Elimination (RFE) with Random Forest (less accurate than removing collinear features)
Model Generation version III: feature selection using Recursive Feature Elimination (RFE) with Logistic Regression (even less accurate than RFE with RF)
Model Prediction with Testing Data
- Load saved four things mentioned above, and do the prediction
More detailed FeatureTools Notes based on the Titanic problem.