osl_dynamics.inference.optimizers

Custom TensorFlow optimizers.

Module Contents

Classes

ExponentialMovingAverage	Optimizer for applying a exponential moving average update.
MarkovStateModelOptimizer	Optimizer for a model containing a hidden state Markov chain.

class osl_dynamics.inference.optimizers.ExponentialMovingAverage(decay=0.1)

Bases: keras.optimizers.optimizer_v2.optimizer_v2.OptimizerV2

Optimizer for applying a exponential moving average update.

Parameters:

decay (float) – Decay for the exponential moving average, which will be calculated as (1-decay) * old + decay * new.

_resource_apply_dense(grad, var)

class osl_dynamics.inference.optimizers.MarkovStateModelOptimizer(ema_optimizer, base_optimizer, learning_rate)

Bases: keras.optimizers.optimizer_v2.optimizer_v2.OptimizerV2

Optimizer for a model containing a hidden state Markov chain.

Parameters:

• ema_optimizer (osl_dynamics.inference.optimizers.ExponentialMovingAverage) – Exponential moving average optimizer for the transition probability matrix.

• base_optimizer (tf.keras.optimizers.Optimizer) – A TensorFlow optimizer for all other trainable model variables.

• learning_rate (float) – Learning rate for the base optimizer.

_create_slots(var_list)

_prepare_local(var_device, var_dtype, apply_state)

_resource_apply_dense(grad, var, **kwargs)

_resource_apply_sparse(*args, **kwargs)