dctools.processing.base.BaseDCEvaluation

class dctools.processing.base.BaseDCEvaluation(arguments)

Base class for evaluation orchestration.

Subclasses are expected to: - define self.dataset_references (pred -> list[ref]) - create self.dataset_processor as needed - implement a run_eval() method

Parameters:: arguments (Namespace)

__init__(arguments)

Parameters:: arguments (Namespace)
Return type:: None

Methods

`__init__`(arguments)
`check_dataloader`(dataloader)	Basic integrity checks on batches.
`close`()	Release resources held by this evaluation run.
`filter_data`(manager, filter_region)	Filter data by time and region.
`get_catalog`(dataset_name, local_catalog_dir, ...)	Ensure a dataset catalog exists locally, downloading if necessary.
`run_eval`()	Run the full evaluation pipeline.
`setup_dataset_manager`(list_all_references)	Build and return a fully configured `MultiSourceDatasetManager`.
`setup_transforms`(dataset_manager, aliases)	Configure and return the transform dict for all aliases.

Attributes

CHALLENGE_NAME

check_dataloader(dataloader)

Basic integrity checks on batches.

Parameters:: dataloader (EvaluationDataloader)
Return type:: None

close()

Release resources held by this evaluation run.

Return type:: None

filter_data(manager, filter_region)

Filter data by time and region.

Parameters:

manager (MultiSourceDatasetManager)
filter_region (Any)

get_catalog(dataset_name, local_catalog_dir, catalog_cfg)

Ensure a dataset catalog exists locally, downloading if necessary.

Parameters:

dataset_name (str)
local_catalog_dir (str)
catalog_cfg (dict)

Return type:

None

run_eval()

Run the full evaluation pipeline.

Return type:: None

setup_dataset_manager(list_all_references)

Build and return a fully configured MultiSourceDatasetManager.

Parameters:: list_all_references (List[str])
Return type:: MultiSourceDatasetManager

setup_transforms(dataset_manager, aliases)

Configure and return the transform dict for all aliases.

When self.args.surface_only is True, every non-GLORYS dataset uses the standardize_to_surface transform (selects the surface depth level only). This is the standard mode for 2-D challenges such as DC1.

Parameters:

dataset_manager (MultiSourceDatasetManager)
aliases (List[str])

Return type:

Dict[str, Any]