This repository contains the source code for our approach to the 2019 WSDM Cup Spotify Sequential Skip Prediction Challenge. We achieved the 5th position under the 'Adrem Data Lab' team. More information on the competition, as well as the dataset, can be found here.
Our solution comprises a modular, scalable and data-efficient pipeline, a minimum of feature engineering and a single recurrent neural network architecture trained with a custom weighted loss function. A detailed motivation and description of our approach can be found in the accompanying paper.
The current script expects a GPU installation of Tensorflow, with Keras on top. Furthermore, model training is parallellised over 2 GPUs.
The main script expects a data-folder containing the gzipped Spotify dataset in training_set, test_set and track_features subfolders, along with empty preprocessed, predictions and submissions subfolders. These latter 3 will hold intermediate (gzipped preprocessed matrices for the training and test set) and output (raw predictions and CrowdAI submission format) files.
The required python packages can be found in requirements.txt.
Using a package manager such as pip, they can easily be installed with the following command:
pip3 install -r requirements.txt
python3 RNN.py <path> <start_day> <end_day> <sub_ID>
<path>: the path to a data-folder containing the Spotify dataset. Intermediate and output files will be written here.<start_day>: the day to start processing on (inclusive, integer between 0 - 65).<end_day>: the day to end processing with (exclusive, integer between 1 - 66).<sub_ID>: Unique identifier for this submission, used to avoid overwriting files between different runs.
