Aggregated metrics manager

fyt.core.aggregated_metrics_manager

AggregatedMetricsManager

Handles aggregation and logging of metrics across multiple seed runs.

This class provides functionality to: - Aggregate metrics from multiple runs - Generate distribution plots with KDE for each metric - Calculate comprehensive statistics (mean, median, std, IQR, min, max, quartiles) - Log results via the experiment logger abstraction

Instances expose input_data (the validated input containing metrics results and experiment name), experiment_name (the name of the experiment), and seeds (the list of seeds extracted from the metrics results), all set during initialization.

__init__(input_data, experiment_logger=None)

Initialize the aggregated metrics manager.

Parameters:

Name	Type	Description	Default
input_data	AggregatedMetricsManagerInput	Validated AggregatedMetricsManagerInput containing metrics and experiment info	required
experiment_logger	BaseExperimentLogger | None	Experiment logger for logging results.	None

generate_aggregated_statistics()

Generate aggregated statistics for all metrics.

Returns:

Type	Description
list[AggregatedMetricResult]	List of AggregatedMetricResult objects

generate_all_plots()

Generate distribution plots for all metrics.

Returns:

Type	Description
dict[str, Figure]	Dictionary mapping metric names to Plotly Figure objects

log_results()

Log aggregated metrics, statistics, and plots via the experiment logger.

Creates a new run under the same experiment and logs: - Aggregated statistics as a dataset (CSV artifact) - Distribution plots for each metric as HTML artifacts - Summary metrics

Returns:

Type	Description
str	The run_id of the created run.