osl_dynamics.inference

TensorFlow layers, callbacks, and utilities used by the models.

This subpackage contains the building blocks that the model classes in osl_dynamics.models use internally. Most users will not need to interact with this subpackage directly.

Modules

• layers — Custom Keras layers (RNN inference networks, softmax layers, sampling layers, observation model layers, etc.).

• callbacks — Training callbacks (KL annealing, transition probability updates, EMA).

• initializers — Weight initialisers for observation model parameters.

• metrics — Loss metrics and evaluation functions.

• modes — Utilities for manipulating inferred mode/state time courses (reordering, matching, correlation).

• optimizers — Custom optimisers.

• regularizers — Regularisers for observation model parameters.

• tf_ops — TensorFlow utility operations.