Trainer

mllabs._trainer.Trainer

Runs cross-validation training on a subset of Pipeline nodes.

Created via :meth:~mllabs.Experimenter.add_trainer. Shares the Experimenter's Pipeline and DataCache.

Attributes:

Name	Type	Description
name	str	Trainer name.
selected_stages	list[str]	Stage nodes included in training.
selected_heads	list[str]	Head nodes to train.
train_folds	list[TrainFold]	Per-split data flows and artifact stores.

select_head(nodes)

Specify Head nodes to train and auto-collect their upstream Stages.

Parameters:

Name	Type	Description	Default
nodes		Node query — list, regex str, or None (all heads).	required

train(n_jobs=1, gpu_id_list=None)

Train all unbuilt selected nodes across all splits.

Stages are trained first (topological order), then Head nodes.

Parameters:

Name	Type	Description	Default
n_jobs	int	Number of parallel workers. Default 1 (sequential).	1
gpu_id_list	list	GPU IDs for GPU-enabled nodes.	None

get_status(node_name)

Return the disk status of a node across all folds.

Returns 'built', 'finalized', 'error', None (init), or 'inconsistent' if folds differ.

get_node_error(node_name)

Return error dict for a node in error state, or None.

process(data, v=None)

Apply trained processors to new data, yielding one result per split.

Parameters:

Name	Type	Description	Default
data		Input dataset.	required
v		Output column filter applied to Head outputs.	None

Yields:

Name	Type	Description
DataFrame		Concatenated Head outputs for each split.

to_inferencer(v=None)

Export trained processors to a standalone :class:~mllabs.Inferencer.

All selected nodes must be in built state.

Parameters:

Name	Type	Description	Default
v		Output column filter passed to the Inferencer.	None

Returns:

Name	Type	Description
Inferencer		Independent inferencer ready for deployment.

Raises:

Type	Description
RuntimeError	If any selected node is not built.

reset_nodes(nodes)

get_n_splits()