Zindi_lightautoml_baseline

This is a repo for AutoInland Vehicle Insurance Claim Challenge baseline based on LightAutoML.

Current score is 0.3252

P.S. We will be happy for your ⭐️ on LightAutoML repo if you like it.

---

Prerequisites

LightAutoML installation from PyPI:

pip install -U lightautoml

To make the code more suitable for your computational resources please change the N_THREADS for the number of threads to train LightAutoML model on your machine.

---

The structure of baseline:

• LightAutoML installation
• Parameters setup
• Data loading
• Feature engineering
• Splitting train dataset into training and validation parts
• Train basic model and:
  • check the score on validation part
  • check feature importances
  • optimize the threshold for F1
• Train timeout utilization model and:
  • check the score on validation part
  • check feature importances
  • optimize the threshold for F1
• Select the best model from upper two and retrain it on the full training dataset
• Prepare submission