Skip to content

Experiments

Iteration

Bases: Protocol

An iteration represents a single step or cycle in the training or evaluation of a model. It typically corresponds to one pass through the training dataset (an step) or a defined unit of model progression.

Iterations can own modules that are associated with the training or evaluation process itself, rather than the model's core architecture.

modules property

An iteration can own a collection of modules not related to the model itself, but to it's training or evaluation process.

Returns:

Name Type Description
Modules Modules

A collection of modules.

step property

The numeric identifier of this iteration in the training or evaluation process. Usually corresponds to the number of passes completed over the dataset.

Returns:

Name Type Description
int Optional[int]

The step number of this iteration.

Iterations

Bases: ABC

A collection of iterations within the lifecycle of a model. Provides an interface for creating, retrieving, and listing iterations.

clear() abstractmethod

Removes all iterations from the collection.

create(step) abstractmethod

Adds a record of an iteration in the database with their respective modules and returns it.

Parameters:

Name Type Description Default
step int

The step number of the iteration to create.

required

get(step) abstractmethod

Retrieves a list of iterations for a given step number.

Parameters:

Name Type Description Default
step int

The step number of the iteration to retrieve.

required

Returns:

Type Description
Optional[Iteration]

list[Iteration]: The iteration list.

list() abstractmethod

Lists all iterations stored in the collection.

Returns:

Type Description
list[Iteration]

list[Iteration]: The complete set of iterations.

remove(iteration) abstractmethod

Removes an iteration from the collection.

Parameters:

Name Type Description Default
iteration Iteration

The iteration to remove.

required