Magenta v0.3.1

Changes:

• Improved Python 3 support

Magenta v0.3.0

Changes:

• Python 3 support
• Pitch class and note density conditioning for Performance RNN

Magenta v0.2.3

Changes:

• Performance RNN release
• Improvements to AI Jam JavaScript demo
• New NSynth notebook example
• Several documentation fixes

Magenta v0.2.0

This release includes the following changes from v0.1.5:

• Add WaveNet fast generator and model to pip package. Thanks to @pkmital!
• Switch to using tf.contrib.training.HParams instead of our own implementation now that it is available in TF v1.1.0. This changes the format of the --hparams flags.
  Example:
  --hparams="{'batch_size': 3, 'rnn_layer_sizes': [128, 128]}'"
  would now be written as
  --hparams="batch_size=3,rnn_layer_sizes=[128,128]".

Magenta v0.1.15

Minor update to add missing python-rtmidi dependency for the AI Jam JavaScript demo.

Magenta v0.1.14

Major changes:

• Added AI Jam JavaScript demo.
• Fixed sequence generator quantization error.

Magenta v0.1.13

Major changes:

• sketch_rnn fixes
• bug fixes and improvements in event_rnn models

Magenta v0.1.12

Major Changes:

• Released sketch_rnn

Magenta v0.1.11

Major Changes:

• Fixed compatibility with newer versions of mido.
• Made Event RNN generation compatible with tuple RNN states.
• Added Pianoroll RNN-NADE model.

Magenta v0.1.10

Major changes:

• TensorFlow 1.0 API compatibility
• Added drums and improv rnn to the pip package