utils¶

load_experiment ¶

load_experiment(exp_path, search_path=None)

Loads all configuration and metrics outputs in an experiment directory.

Parameters:

Name	Type	Description	Default
`exp_path`	`str \| Path`	The directory to search for data. Directory must include the ".hydra/config.yaml" file.	required

Returns:

Name	Type	Description
`exps`	`Union[Experiment, List[Experiment]]`

load_from_checkpoint ¶

load_from_checkpoint(
    model,
    *,
    ckpt=None,
    weights_key=None,
    weights_key_strip=None,
    model_attr=None,
)

Load model weights.

Parameters:

Name	Type	Description	Default
`model`	`Module`	The PyTorch Module	required
`ckpt`	`Optional[Union[str, Path]]`	The path to the file containing the model weights. If no path is provided the model will not be updated.	`None`
`weights_key`	`Optional[str] (default: None)`	The key from the checkpoint file containing the model weights. When None, the checkpoint dict is loaded into `model` directly.	`None`
`weights_key_strip`	`Optional[str] (default: None)`	A prefix to remove from each weight's key prior to loading. When None, keys are used as-is.	`None`
`model_attr`	`Optional[str] (default: None)`	The attribute of `model` to load the weights into. When None, they are loaded into `model` itself.	`None`

Returns:

Name	Type	Description
`model`	`Module`	The same `model` instance, with its weights loaded.

original_cwd ¶

original_cwd()

Directory the experiment was launched from.

Inside a Hydra job the process cwd is the per-job output directory, so relative paths in a task() silently resolve against the wrong place. Use this to anchor paths against the launch directory instead::

data = load(mushin.original_cwd() / "data" / "train.csv")

Outside a Hydra run this is just the current working directory.