drytorch.core.experimenting

Module containing the Experiment and Run class.

Classes

`Experiment`(config, *[, name, par_dir, tags])	Manage experiment configuration, directory, and tracking.
`Run`(experiment, run_id[, resumed, record])	Execution lifecycle for a single run of an Experiment.
`RunMetadata`(id, status, timestamp, commit)	Metadata for a run.
`RunRegistry`(path)	Creates and manages a JSON file for run metadata.

class Experiment(config: _T_co, *, name: str = '', par_dir: str | Path = PosixPath('.'), tags: list[str] | None = None)[source]

Bases: Generic[_T_co]

Manage experiment configuration, directory, and tracking.

This class associates a configuration file, a name, and a working directory with a machine learning experiment. It also contains the trackers responsible for tracking the metadata and metrics for the experiment. Finally, it allows global access to a configuration file with the correct type annotations.

folder_name

name of the hidden folder storing experiment metadata.

Type:: ClassVar[str]

run_file

filename storing the registry of run IDs for this experiment.

Type:: ClassVar[str]

previous_runs

a list of all previous runs created by this class.

Type:: ClassVar[list[drytorch.core.experimenting.Run]]

par_dir

parent directory for experiment data.

Type:: pathlib.Path

tags

descriptors for the experiment.

Type:: list[str]

trackers

dispatcher for publishing events.

Type:: drytorch.core.tracking.EventDispatcher

Initialize.

Parameters:

config (_T_co) – Configuration for the experiment.
name (str) – The name of the experiment (defaults to class name).
par_dir (Path) – Parent directory for experiment data.
tags (list[str]) – Descriptors for the experiment (e.g., "lr=0.01").

property name: str: The name of the experiment.

property config: _T_co: Experiment configuration.

create_run(*, run_id: str | None = None, resume: bool = False, record: bool = True) → Run[_T_co][source]

Convenience constructor for a Run using this experiment.

Parameters:

run_id (str | None) – identifier of the run; defaults to timestamp.
resume (bool) – resume the selected run if run_id is set, else the last run.
record (bool) – record the run in the registry.

Returns:

The created run object.

Raises:

RunAlreadyRecordedError – if creating a new run with an existing id.

Return type:

Run[_T_co]

property run: Run[_T_co]

Get the current run.

Raises:: NoActiveExperimentError – if no run is currently active.

classmethod get_config() → _T_co[source]

Retrieve the configuration of the current experiment.

Return type:: _T_co

classmethod get_current() → Self[source]

Return the currently active experiment.

Raises:: NoActiveExperimentError – if no experiment is currently active.
Return type:: Self

static set_current(experiment: Experiment[_T_co]) → None[source]

Set an experiment as active.

Raises:: NestedScopeError – if there is an already active run.
Parameters:: experiment (Experiment[_T_co])
Return type:: None

class Run(experiment: Experiment[_T_co], run_id: str | None, resumed: bool = False, record: bool = True)[source]

Bases: CreatedAtMixin, Generic[_T_co]

Execution lifecycle for a single run of an Experiment.

status

Current status of the run.

Type:: RunStatus

resumed

whether the run was resumed.

Type:: bool

metadata_manager

Manager for run metadata.

Type:: tracking.MetadataManager

record

whether to record the run in the registry.

Type:: bool

Initialize.

Parameters:

experiment (Experiment[_T_co]) – the experiment this run belongs to.
run_id (str | None) – identifier of the run.
resumed (bool) – whether the run was resumed.
record (bool) – record the run in the registry.

property experiment: Experiment[_T_co]: The experiment this run belongs to.

property id: str: The identifier of the run.

__enter__() → Self[source]

Enter the experiment scope.

Return type:: Self

__exit__(exc_type: type[BaseException] | None, exc_val: BaseException | None, exc_tb: TracebackType | None) → None[source]

Exit the experiment scope.

Parameters:

exc_type (type[BaseException] | None)
exc_val (BaseException | None)
exc_tb (TracebackType | None)

Return type:

None

is_active() → bool[source]

Check if the run is currently active.

Return type:: bool

stop() → None[source]

Stop the experiment scope.

Return type:: None

start() → None[source]

Start the experiment scope.

Parameters:: self (Self)
Return type:: None