opto.optimizers.optimizer

Module Contents

Classes

AbstractOptimizer	An optimizer is responsible for updating the parameters based on the feedback.
Optimizer

API

class AbstractOptimizer(parameters: List[opto.trace.nodes.ParameterNode], *args, **kwargs)

An optimizer is responsible for updating the parameters based on the feedback.

Initialization

abstract step()

Update the parameters based on the feedback.

abstract zero_feedback()

Reset the feedback.

abstract property propagator

Return a Propagator object that can be used to propagate feedback in backward.

class Optimizer(parameters: List[opto.trace.nodes.ParameterNode], *args, propagator: opto.trace.propagators.propagators.Propagator = None, **kwargs)

Bases: opto.optimizers.optimizer.AbstractOptimizer

property propagator

property trace_graph

Aggregate the graphs of all the parameters.

step(*args, **kwargs)

propose(*args, **kwargs)

Propose the new data of the parameters based on the feedback.

update(update_dict: Dict[opto.trace.nodes.ParameterNode, Any])

Update the trainable parameters given a dictionary of new data.

zero_feedback()

default_propagator()

Return the default Propagator object of the optimizer.

backward(node: opto.trace.nodes.Node, *args, **kwargs)

Propagate the feedback backward.