EB RecSys Challenge

Results

Method	AUC ↑	MRR ↑	nDCG@5 ↑	nDCG@10 ↑
Random (small)	0.4998	0.3156	0.3489	0.4338
NRMS (small)	0.5299	0.3243	0.3625	0.4420
GERL (small)	*0.5820*	*0.3587*	*0.4032*	*0.4775*

Method	AUC ↑	MRR ↑	nDCG@5 ↑	nDCG@10 ↑
GERL (demo)	*0.5344*	*0.3231*	*0.3610*	*0.4436*
GERL (demo) + weighted sampling	0.5258	0.3172	0.3544	0.4377

Create a virtual environment and install the needed dependencies:

python -m venv .venv
. .venv/bin/activate
pip install -r requirements.lock

Or alternatively, use Rye:

rye sync
. .venv/bin/activate

From the root of the repository, run the preprocessing script:

python -m src.recsys_challenge.dataset.preprocess.auto

To train the model, run:

python -m src.recsys_challenge.training

The default options for training are set to the params we used for training. Options for the training script can be printed using:

python -m src.recsys_challenge.training --help