/RL_Lux_AI

Primary LanguageJupyter Notebook

Run.ipynb File can be used to train the models and create submission file.

Train a models for Ensemble method

# Model parameters could be changed from line 334
# update ./UNet_attention_jason_0-3_ensemble/train.py
# For each epoch, model is saved if current epoch shows higher accuracy than previous epoch.
# Early stopping is applied so if model does not see improvement for 3 epochs, the training will stop.

!python UNet_attention_jason_0-3_ensemble/train.py

Create Ensemble submission file

# Submission file will be created at ./UNet_attention_jason_0-3

%cd f'./UNet_attention_jason_0-3''
!tar -czf submission.tar.gz *

Modify Ensemble Model

  • If you have new model files (.pth) that you would like to include, modify ./UNet_attention_jason_0-3_ensemble/agent.py
  • Note that Kaggle submission file has to be less than 100Mb

Train/Modify Transformer method

# Open the file transformer/Transformer_LuxAI.ipynb
# Follow the instructions inside. You can tune your hyperparameter inside this notebook.
# This notebook will save model.pth from each epoch. Simply observe the accuracy after each epoch from choosing the model you want.

Files greater than 100mb has been removed in accordance to github policy. Removed files:

  • ./UNet_attention_jason_0-3_ensemble/submission.tar
  • ./transformer/checkpoint.pth