4th place solution (tattaka's part)

Environment

Use Kaggle Docker.

Usage

0. Place competition data in the input directory

1. make bp_matrix

   $ cd src/scripts
   $ python make_structure_feat_train.py
   $ python make_structure_feat_test.py

   Place contrafold bpps and unzip in in the input directory

   • https://www.kaggle.com/code/tattaka/contrafold-test-bpps-1000000
   • https://www.kaggle.com/code/tattaka/contrafold-test-bpps-500000-1000000
   • https://www.kaggle.com/code/tattaka/contrafold-test-bpps-0-500000
   • https://www.kaggle.com/code/tattaka/contrafold-train-bpps-600000
   • https://www.kaggle.com/code/tattaka/contrafold-train-bpps-300000-600000
   • https://www.kaggle.com/code/tattaka/contrafold-train-bpps-0-300000

2. Training

   $ cd src/exp064 && sh train.sh
   $ cd src/exp070 && sh train.sh
   $ cd src/exp071 && sh train.sh 
   $ cd src/exp072 && sh train.sh

License

MIT