Experiment

lit.sdk.model.experiment

This module provides functionality for managing and controlling experiments. An experiment represents a model's training session and includes methods for starting, stopping, resuming, and querying experiments.

Experiment

Represents a model's experiment, or training session.

is_running = self.meta.get('is_running', False) instance-attribute

Whether the experiment is currently running.

meta = meta or {} instance-attribute

Additional metadata.

runid = runid instance-attribute

The index of the run.

team = team instance-attribute

The name of a team.

__init__(team, runid, meta=None)

Initializes a new instance of Experiment.

Parameters:

Name	Type	Description	Default
team	str	The team which owns the Experiment.	required
runid	int	The run id of the Experiment.	required
meta	dict | None	Extra optional metadata for the Experiment. Defaults to None.	None

resume(device, smartfit=False, session_name=None)

Resumes a stopped experiment.

Parameters:

Name	Type	Description	Default
device	string	The name of a device; e.g. gpu4	required
session_name	string | None	The name of the previously running experiment. Defaults to None.	None

Raises:

Type	Description
LITModelException	If an error occurred while resuming the experiment.

Returns:

Type	Description
str	The name of the resumed experiment session.

stop()

Stops the running experiment.

Raises:

Type	Description
LITModelException	If an error occurred while stopping the experiment.

Returns:

Type	Description
int	The exit code from the killed process.

get_experiments(team)

Get a list of all experiments, running and stopped.

Parameters:

Name	Type	Description	Default
team	str	The name of the team.	required

Raises:

Type	Description
LITModelException	If an error occurred while getting experiments.

Returns:

Type	Description
list[Experiment]	A list of experiment objects.

Examples:

>>> experiments = get_experiments("contoso")
>>> experiments[0].runid, experiments[0].is_running
(73, False)

halt_experiments(team)

Immediately halts all running experiments for a given team.

Parameters:

Name	Type	Description	Default
team	str	The name of the team.	required

Raises:

Type	Description
LITModelException	If an error occurred while halting all experiments.

start_experiment(team, name, device, smartfit=False)

Parameters:

Name	Type	Description	Default
team	str	The name of the team.	required
name	str	The name of the canvas.	required
device	str	The name of device. i.e. gpu0, cpu	required

Raises:

Type	Description
LitModelError	If an error occurred while starting the experiment.

Returns:

Type	Description
str	The name of the training session.