2025 Mitsui Commodity Price Prediction

How to run

Run the script with

cd src
python study.py

If you want, you can override configs:

python study.py model=lstm experiment.optuna.n_trials=20

You can try different models:

python study.py model=transformer

For managing multiple model architectures:

1. Add configs like configs/model/transformer.yaml
2. Update MODEL_CLASSES and TRAINER_CLASSES in study.py
3. Run separate studies for each model:

python study.py model=lstm experiment.optuna.study_name=lstm_hpo
python study.py model=transformer experiment.optuna.study_name=transformer_hpo

Start testing with small subset

Choose 10% of rows and 20 targets(out of 400+) to balance speed and representativeness.

Validate Configs:

python -c "from utils import validate_all_configs; validate_all_configs('configs/model/')"

Test Subset:

python src/start_small.py

Run Full Pipeline:

cd src 
python study.py -m model=lstm, cnnlstm, tcn, fed