My code for Telstra Network Disruptions Kaggle competition.
This code has a companion blog post with my competition writeup.
Competition data can be downloaded here and should go into the data folder. data folder also contains my final ensemble's predictions for the test set and out-of-fold predictions for the train set.
See my notebooks for:
- Automatic model tuning with Sacred and Hyperopt
- Discovering the magic feature and some visualizations
- Neural Net and Xgboost models and my blending approach
- Global Refinement of Random Forest
To be added later:
- Calibrating probabilities
Code for loading data and building features is in src/telstra_data.py.