Requirements

• torch == 1.1.0
• cnn_finetune == 0.5.3
• albumentations == 0.2.3
• catalyst == 19.06.5

How to train

Pretrained with controls

bash bin/train_control.sh

Pretrained models are saved at: /raid/bac/kaggle/logs/recursion_cell/pretrained_controls/$channels/se_resnext50_32x4d/
where channels can be: [1,2,3,4,5], etc

Train with pseudo data

bash bin/train_pseudo.sh

• PRETRAINED_CONTROL: is the root folder of pretrained models above

• --model_params/weight= $PRETRAINED_CONTROL/$channels/se_resnext50_32x4d/checkpoints/best.pth:str \ is the weight of corresponding model pretrained on controls dataset.

Train with usually data

Similar to Train with pseudo data part

bash bin/train.sh

• PRETRAINED_CONTROL: is the root folder of pretrained models above

• --model_params/weight= $PRETRAINED_CONTROL/$channels/se_resnext50_32x4d/checkpoints/best.pth:str \ is the weight of corresponding model pretrained on controls dataset.

• Run: python src/make_submission.csv