model.runtime.workflows.temporal_cv_workflow

Temporal cross-validation workflow for ADELM calibration.

Head-agnostic: it folds the calibration window into contiguous time blocks (holding out one block per fold, all entities kept) and delegates the per-fold train / evaluate / save to the runner hooks, so the same workflow drives the site head (flux towers) or the gauge head (gauges) unchanged.

Module Contents

Classes

TemporalCVWorkflow

Run calibration under temporal (by-time-block) cross-validation.

API

class model.runtime.workflows.temporal_cv_workflow.TemporalCVWorkflow(config, config_path=None, config_overrides=None)

Bases: model.runtime.workflows.base_workflow.BaseWorkflow

Run calibration under temporal (by-time-block) cross-validation.

Initialization

workflow_name

‘temporal_cv’

run(epochs=10, lr=0.001, train_chunk_size=None, max_grad_norm=1.0, skip_nan_grads=True, show_progress=True, debug=False, output_dir=None, fold_id=0)