Skip to content

PyTorch implementation of Artist Group Factors (arXiv:1805.02043) and GradNorm (arXiv:1711.02257)

Notifications You must be signed in to change notification settings

falkaer/artist-group-factors

Repository files navigation

This project is an attempt to replicate the findings presented in Transfer Learning of Artist Group Factors to Musical Genre Classification.

This code was written to run with the medium Free Music Archive dataset, which must be extracted to the working directory as the code.

Features and AGFs are extracted from the dataset by running extractor.py and agf.py, and multi-task and single-task models can be trained by running train_mtn.py and train_predictor.py, respectively.

Model architectures can be seen in models.py, and the GradNorm implementation can be seen in the train_mtn.py step method.

predict_slices.py makes predicitons for the last layer of the single-task network, which are used for visualising the internal state of the network before prediction using t-SNE.

The utils.py, tracks.csv and genres.csv files in this repository are from the FMA repository.

About

PyTorch implementation of Artist Group Factors (arXiv:1805.02043) and GradNorm (arXiv:1711.02257)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages