python main.py --common_config <common_config_name> --model <model_name> --model_config <config_name> --dataset <dataset_name> --fold <fold_number> --mode <mode>

Example:

python main.py --common_config config1 --model rf --model_config config1 --dataset pa_lov --fold 0 --mode fit_predict

One can define such commands in a shell script and run them in parallel. For example,

# measure time
start=`date +%s`

python main.py --model rf --dataset pa_lov --config config1 --mode fit_predict --fold 0
python main.py --model rf --dataset pa_lov --config config1 --mode fit_predict --fold 1
python main.py --model rf --dataset pa_lov --config config1 --mode fit_predict --fold 2
python main.py --model rf --dataset pa_lov --config config1 --mode fit_predict --fold 3
python main.py --model rf --dataset pa_lov --config config1 --mode fit_predict --fold 4

# measure time
end=`date +%s`

runtime=$((end-start))

# print time in miniutes upto 2 decimal places
echo "Total time : $(echo "scale=2; $runtime/60" | bc -l) minutes"

Note that sh files are by default .gitignored. So, you can create any custom run.sh file in the root of the repo and run it.

• [To Add] A notebook to prepare the data which goes into the pipeline. Assigned to Z in one of the issues.
• Use notebooks/prepare_lov.ipynb to prepare the data
• Then use main.py to run the experiments

kdd24
|---models          # for all models
    |---<model1_name>
        |---model.py
        |---config1.toml
        |---config2.toml
        |---...
    |---<model2_name>
        |---model.py
        |---config1.toml
        |---config2.toml
        |---...
    ...
|---datasets        # for all datasets
    |---<dataset1_name>
        |---dataset.py
    |---<dataset2_name>
        |---dataset.py
    ...
main.py             # for running experiments

def fit(train_data, config):
    # Fit and save model and auxillary information

def predict(test_data, train_data, config):
    # Predict and save predictions

def fit_predict(train_data, test_data, config):
    # Fit and predict and save predictions. Useful for small models whis do not take much time to fit. For other models, one can use fit and predict separately.

• Open VSCode in the root of the repo

• Install this repo as an editable package

  pip install -e .

• Define your own model in models/<model_name>/model.py which should have fit, predict and fit_predict functions.

• All models must receive the data in same format.

• All models must save their results in the same format.