Models

Generators

Models

The following files define the different model types you can create for generators:

• generator_gru.py
• generator_lstm.py
• generator_transformer.py

Checkpoints

There are subdirectories for each tokenization of the dataset used (naive, miditok, miditok augmented). The trained model weights (checkpoints) are stored in these.

Generator Factory:

Allows you to easily create different types of generators (options are lstm, gru, and transformer). The factory returns a generator model instance.

Arguments for get_generator are:

• model_type (lstm, gru, transformer)
• vocab_size
• kwargs (other architecture specific params):
  • common kwargs:
    • embed_size
    • hidden_size
    • num_layers
    • dropout
  • for transformers:
    • d_model (model dimension)
    • nhead (number of attention heads)
    • dim_feedforward (feedforward dimension)
    • max_seq_length

get_default_config gets the default hyperparams for a given model type.

Discriminators

Models

The following files define the different model types you can create for discriminators:

• discriminators_lstm.py
• discriminators_mlp.py
• discriminator_transformer.py

Checkpoints

There are subdirectories for each tokenization of the dataset used (naive, miditok, miditok augmented). The trained model weights (checkpoints) are stored in these.

Discriminator Factory:

Allows you to easily create different types of discriminators (options are lstm, mlp, and transformer). The factory returns a discriminator model instance.

Arguments for get_discriminator are:

• model_type (lstm, mlp, transformer)
• hidden1 and hidden2 (size of first and second hidden layer)
• pitch_dim
• context_measures
• hidden1 and hidden2: size of hidden layers
• pool
• dropout
• lstm and mlp
  • embed_size
  • hidden_size
  • num_layers
  • dropout
• transformer
  • embed_size
  • num_heads
  • num_layers

get_default_config gets the default hyperparams for a given model type.