Lukass/introduce test mode

[SzymonLukasik]

• train_and_validate refactor
• demo notebook with training and test run from command line

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[adambiel22]

I imagined that we add 3 subparsers.

1. validate to do cross validation, so the process that was done previously in train_and_validate.py.
   • input: method, config
   • output: metrics, and trained model weights (if there was set the option retrain)
2. train to do only training on the whole dataset. So only the last step that was done during cross validation.
   • input: method, config
   • output: trained model weights, and according to task formulation: a table where each cell will be assigned cell-type probabilities (if available) and a final cell-type prediction
3. test / predict
   • input: method, pretrained model weights
   • output: metrics, and according to task formulation: a table where each cell will be assigned cell-type probabilities (if available) and a final cell-type prediction

[adambiel22]

Because to fulfill task requirements 'pre-train baselines and state-of-the-art methods on new data' we don't need cross validation