osl_dynamics.config_api.wrappers.train_hive#

osl_dynamics.config_api.wrappers.train_hive(data, output_dir, config_kwargs, init_kwargs=None, fit_kwargs=None, save_inf_params=True)#

Train a HIVE Model.

This function will:

  1. Build an hive.Model object.

  2. Initialize the parameters of the HIVE model using the method set by init_method in the config (Model.initialization).

  3. Perform full training.

  4. Save the inferred parameters (state probabilities, means, covariances and embeddings) if save_inf_params=True.

This function will create two directories:

  • <output_dir>/model, which contains the trained model.

  • <output_dir>/inf_params, which contains the inferred parameters. This directory is only created if save_inf_params=True.

Parameters:
dataosl_dynamics.data.Data

Data object for training the model.

output_dirstr

Path to output directory.

config_kwargsdict

Keyword arguments to pass to osl_dynamics.models.hive.Config. Defaults to:

{
    'sequence_length': 200,
    'spatial_embeddings_dim': 2,
    'dev_n_layers': 5,
    'dev_n_units': 32,
    'dev_activation': 'tanh',
    'dev_normalization': 'layer',
    'dev_regularizer': 'l1',
    'dev_regularizer_factor': 10,
    'batch_size': 128,
    'learning_rate': 0.005,
    'lr_decay': 0.1,
    'n_epochs': 30,
    'do_kl_annealing': True,
    'kl_annealing_curve': 'tanh',
    'kl_annealing_sharpness': 10,
    'n_kl_annealing_epochs': 15,
}.
init_kwargsdict, optional

Keyword arguments to pass to the initialization method (Model.initialization). Defaults to:

{'n_init': 10, 'n_epochs': 2}.
fit_kwargsdict, optional

Keyword arguments to pass to the Model.fit. No defaults.

save_inf_paramsbool, optional

Should we save the inferred parameters?