kaggle-severstal

https://www.kaggle.com/c/severstal-steel-defect-detection

Catalyst-solution

1. Install mlcomp with pip install mlcomp and pip install pytorch==1.2.0
2. Run mlcomp-server start
3. Run mlcomp init
4. Move all competition data to ~/mlcomp/data/severstal/input/
5. Make sure to have all read\write permissions for data.
6. To apply config to server, run mlcomp dag *path to dir*/kaggle.yml
7. Final model is a trace.pth file in ~/mlcomp/tasks/*number of the task*/trace.pth. It is made in the end of task execution so plan your training time carefully.

Note: Make sure to check avalibility of the model defined by catalyst config with\without grid params!