• 2019 ICLR, Cheng-Zhi Anna Huang, Google Brain
• Re-producer : Jason Yang ( Yang-Kichang )
• paper link
• paper review

1. This Repository is perfectly cometible with tensorflow 2.0

• Domain: Dramatically reduces the memory footprint, allowing it to scale to musical sequences on the order of minutes.

• Algorithm: Reduced space complexity of Transformer from O(N^2D) to O(ND).

• In this repository using single track method (2nd method in paper.).

• If you want to get implementation of method 1, see here .

• I refered preprocess code from performaceRNN re-built repository..

• Preprocess implementation repository is here.

  

• learning rate : 0.0001
• head size : 4
• number of layers : 6
• seqence length : 2048
• embedding dim : 256 (dh = 256 / 4 = 64)
• batch size : 3

• mt.generate() can generate music automatically.

from models import MusicTransformer
mt = MusicTransformer(
  	embedding_dim=256, vocab_size=par.vocab_size, 
  	num_layer=6, 
  	max_seq=max_seq,
  	dropout=0.1,
  	debug=False
)
mt.generate(prior=[1,3,4,5], length=2048)

you can't use tf.keras directly in alpha ver. So you should import from tensorflow.python import keras ,then use > keras.{METHODS}

• example :

  from tensorflow.python import keras 
  dropout = keras.layers.Dropout(0.3)

tf-2.0alpha currently not supported keras.optimizers as version 2. so, you can't use optimizer.apply_gradients(). So, you should import from tensorflow.python.keras.optimizer_v2.adam import Adam first.

• example:

  from tensorflow.python.keras.optimizer_v2.adam import Adam
  optimizer = Adam(0.0001)

current tf 2.0(alpha) , subclassed keras model can't use method save(), summary(), fit() and save_weigths() with .h5 format