nirs4all.visualization.chart_utils.predictions_adapter module

PredictionsAdapter - Adapter for Predictions API with optimized data access.

Wraps the refactored Predictions API to provide convenient methods for charts.

class nirs4all.visualization.chart_utils.predictions_adapter.PredictionsAdapter(predictions)

Bases: object

Adapter for Predictions API with optimized data access.

Wraps the refactored Predictions API to provide convenient methods for charts. Leverages predictions.top(), lazy loading, and structured results.

Key Optimizations: - Uses predictions.top() for efficient ranking - Supports lazy loading (load_arrays=False) for metadata-only queries - Works with PredictionResult/PredictionResultsList classes - Avoids redundant metric calculations

predictions

Predictions object instance.

extract_metric_values(predictions_list: PredictionResultsList, metric: str, partition: str = 'test') → List[float]

Extract metric values from prediction results.

Parameters:

• predictions_list – List of prediction results.

• metric – Metric name to extract.

• partition – Partition to extract from (default: ‘test’).

Returns:

List of metric values.

get_all_predictions_metadata(rank_metric: str = 'rmse', rank_partition: str = 'test', **filters) → PredictionResultsList

Get all predictions matching filters (metadata only, fast).

Parameters:

• rank_metric – Metric for sorting (default: ‘rmse’).

• rank_partition – Partition for sorting (default: ‘test’).

• **filters – Filters to apply (dataset_name, model_name, etc.).

Returns:

PredictionResultsList with all matching predictions (no arrays loaded).

get_top_models(n: int, rank_metric: str, rank_partition: str = 'val', ascending: bool | None = None, load_arrays: bool = True, **filters) → PredictionResultsList

Get top N models using predictions.top() API.

Parameters:

• n – Number of top models to retrieve.

• rank_metric – Metric to rank by.

• rank_partition – Partition to rank on (default: ‘val’).

• ascending – Sort order (None = auto-detect from metric).

• load_arrays – Whether to load prediction arrays (default: True).

• **filters – Additional filters (dataset_name, model_name, etc.).

Returns:

PredictionResultsList of top N models.

static is_higher_better(metric: str) → bool

Check if metric is higher-is-better.

Parameters:

metric – Metric name.

Returns:

True if higher is better, False otherwise.