Attribute a song to an author or a genre
Our project depends on Hedwig, which is designed for Python 3.6 and PyTorch 0.4. PyTorch recommends Anaconda for managing your environment. We'd recommend creating a custom environment as follows:
$ conda create --f environment.yml
Code depends on data from NLTK (e.g., stopwords) so you'll have to download them. Run:
$ python src/nltk_download.py
Download the Reuters, word2vec embeddings from hedwig-data.
$ git clone https://github.com/j-cahill/hedwig.git
$ git clone https://git.uwaterloo.ca/jimmylin/hedwig-data.git
After cloning the hedwig-data repo, you need to unzip the embeddings and run the preprocessing script:
cd hedwig-data/embeddings/word2vec
gzip -d GoogleNews-vectors-negative300.bin.gz
python bin2txt.py GoogleNews-vectors-negative300.bin GoogleNews-vectors-negative300.txt
Download 380,000 lyrics from metrolyrics and 55,000+ song lyrics from Kaggle. Run
$ python src/data_preprocessing.py
To impute the missing genre values using the Genius API. Then run
$ python src/lyrics_preprocessor.py
$ python src/lyrics_preprocessor_artists.py
to perform basic data cleanup and create train/test/dev splits for both genre and artist.
It is imperative for the code to run that the repo be structured in the following way
Deep-Learning-Project
│
|---hedwig
│
└---hedwig-data
│----LyricsGenre
| train.tsv
| test.tsv
| dev.tsv
│----LyricsArtist
| train.tsv
| test.tsv
| dev.tsv
Other files and folder may be present as well but this structure must be observed.
Train a BERT model with the following command, testing will immediately follow
$ cd hedwig/
$ python -m models.bert --dataset LyricsGenre --model bert-base-uncased --max-seq-length 256 --batch-size 16 --lr 2e-5 --epochs 2
$ cd hedwig/
$ python -m models.bert --dataset LyricsArtist --model bert-base-uncased --max-seq-length 256 --batch-size 16 --lr 2e-5 --epochs 20
Models are saved to hedwig/model_checkpoints