cellmil.utils.wandb

Classes

ExperimentComponents(task, features, model, ...)	Parsed components of an experiment ID.
WandbClient(team, projects[, tasks])	Client for retrieving and processing wandb runs.

class cellmil.utils.wandb.ExperimentComponents(task: str, features: str, model: str, regularization: str, stratification: str)

Bases: object

Parsed components of an experiment ID.

task: str

features: str

model: str

regularization: str

stratification: str

property has_regularization: bool

property has_stratification: bool

__init__(task: str, features: str, model: str, regularization: str, stratification: str) → None

class cellmil.utils.wandb.WandbClient(team: str, projects: list[str], tasks: Optional[list[str]] = None)

Bases: object

Client for retrieving and processing wandb runs.

__init__(team: str, projects: list[str], tasks: Optional[list[str]] = None)

Initialize the WandB client.

Parameters:

• team – Team name where the projects belong in wandb.

• projects – List of project names to retrieve runs from.

• tasks – Optional list of valid tasks for filtering runs.

get_runs(preprocess: bool = True) → list[Any]

Retrieve wandb runs for configured projects and team.

Parameters:

preprocess – Whether to preprocess the runs (default: True).

Returns:

List of wandb runs (preprocessed if requested).

get_experiment_id(run: Any) → str

Get the experiment ID from a run.

Handles both formats: - FOLD_N_EXPERIMENTID_YYYY-MM-DD_HH-MM-SS - FINAL_EXPERIMENTID_YYYY-MM-DD_HH-MM-SS

Parameters:

run – The wandb run object

Returns:

The extracted experiment ID

static parse_experiment_components(experiment_id: str) → ExperimentComponents

Parse an experiment ID into its components.

Format: TASK+FEATURES+MODEL+REG+STRA Example: DCR+ALL+ABMIL+REG+* or OS+PYRAD+HEAD4TYPE+*+STRA

Parameters:

experiment_id – The experiment ID string

Returns:

ExperimentComponents with parsed values

Raises:

ValueError – If experiment ID doesn’t match expected format

get_task(experiment_id: str) → str | None

Get the task associated with a given experiment ID.

Parameters:

experiment_id – The ID of the experiment

Returns:

The name of the task, or None if not valid

Raises:

ValueError – If tasks are specified and the experiment ID doesn’t correspond to a known task

_has_dataloader_error(run: Any, exp_id: str) → bool

Check if a run has a DataLoader worker error in its logs.

Parameters:

• run – A wandb run object

• exp_id – The experiment ID to look for in the error message

Returns:

True if the error is found, False otherwise

static get_metric(run: Any, metric: str) → float

Get the highest validation metric across all epochs for a given run.

Parameters:

• run – A wandb run object

• metric – The metric name (e.g., “f1”, “c_index”, “balacc”)

Returns:

The highest validation metric score

Raises:

ValueError – If the metric is not found or has no valid values