version.txt

Version Updates

v.0.2.0 - 06/10/18
- added custom loss function
- velocities are now normalized
- output is still a bit buggy (low probs + velocities produced)

v.0.1.1 - 06/03/18
A few changes:
- changed the model syntax in polyphonic_lstm_training.py to functional API
- changed generation.py to handle the new model outputs (probs/vels)
changes to consider:
- cost function: currently loss is mean squared on *both probabilities and velocities*
- maybe there's a better way to unpack the model output array (length 2*new_num_notes) into the probs and vels arrays (length new_num_notes for both)

v.0.1.0 - 05/31/18
- changed models to produce probabilities and velocities

v.0.0.2 - 05/26/18
- mostly done accommodating pianoroll data type
  - model inputs are now pianoroll files (i.e. arrays where nonzero entries are not necessarily 1)
- issue to resolve: output still binary (i.e. probability vector is outputted, with notes sampled from prob vector. How to add velocities to output?)
  - relatedly, what is the best cost to use?

v.0.0.1 - 05/21/18
- changed data_processing to produce pianoroll files without normalizing velocities

v.0.0.0 - 05/21/18
- Forked from brunnergino/JamBot