- Download datasets from here
- Setup environment with
conda env create -f environment.yml - Train a baseline MTL system with
python train.py --do-train --eval-every 2000 --run-name baseline - Evaluate the system on test set with
python train.py --do-eval --sub-file mtl_submission.csv --save-dir save/baseline-01 - Upload the csv file in
save/baseline-01to the test leaderboard. For the validation leaderboard, runpython train.py --do-eval --sub-file mtl_submission_val.csv --save-dir save/baseline-01 --eval-dir datasets/oodomain_val