V0.5 Model Weights

14,0.983495887296,0.266003160914,0.70466523522,3.3920778999

bytehoven-66-weights -- the model weights distributed with this release allow Bytehoven to distinguish between
trainingresults-66.log
66 composers.
composerlist-66.csv -- composer list in order network outputs
trainingresults-66.log -- log file from model fitting, optimal fit epoch 14 (of 9th model iteration). Total training time about 96 hours.

Accuracy training set: 98.35%
Accuracy dev set: 70.47%
Accuracy test set: 69.44%

V0.4 Model Weights

bytehoven-45-weights -- the model weights distributed with this release allow Bytehoven to distinguish between 45 composers.
composerlist-45.csv -- composer list in order network outputs
trainingresults-45.log -- log file from model fitting, optimal fit epoch 13 (of 3rd model iteration). Total training time about 48 hours.

The dataset is not included, it has become a bit unwieldy in size (12.7GB). The final dataset will be published. Send me a message if you feel you cannot wait for that and I'll arrange a transfer.

Accuracy training set: 99.97%
Accuracy dev set: 71.30%
Accuracy test set: 73.53%

V0.2 Model Weights

bytehoven-22-weights -- the model weights distributed with this release allow Bytehoven to distinguish between 22 composers.
composers.csv -- composer list in order network outputs
trainingresults-22.log -- log file from model fitting, optimal fit epoch 61

Musicdata_Medium_22composers -- dataset used in training and evaluation (increments V0.1, please download both and extract to the same folder)

Accuracy training set: 98.23%
Accuracy dev set: 73.41%
Accuracy test set: 76.44%

V0.1 model weights

bytehoven-7-weights -- the model weights distributed with this release allow Bytehoven to distinguish between 7 composers: Bach, Beethoven, Brahms, Chopin, Grieg, Liszt, and Mozart.

trainingresults.log -- log file from model fitting

Musicdata_Medium_7composers -- dataset used in training and evaluation