ctf4science.eval_module

Evaluation Module for CTF models, provides evaluation metrics and routines for CTF datasets.

This module handles evaluation of CTF models against a hidden test set. It also assesses model stability by running models multiple times with different random seeds.

Functions

compute_log_psd(array, k, modes)	Compute the natural log of the averaged PSD over the last k time steps.
compute_psd(array, k, modes)	Compute the averaged power spectral density over the last k time steps for the given modes.
evaluate(dataset_name, pair_id, prediction)	Evaluate the prediction using specified metrics; ground truth is loaded internally.
evaluate_custom(dataset_name, pair_id, ...)	Evaluate the prediction against a provided truth array using specified metrics.
evaluate_kaggle_csv(csv_path, dataset_name)	Evaluate the predictions from a Kaggle CSV file.
extract_metrics_in_order(dataset_name, ...)	Extract metric values from batch results in the order defined by the dataset config.
long_time_forecast_dynamical(truth, ...)	Compute the long-time forecast score for dynamical systems (histogram-based).
long_time_forecast_spatio_temporal(truth, ...)	Compute the long-time forecast score for spatio-temporal systems (spectral).
reconstruction(truth, prediction)	Compute the reconstruction score (relative L2 over full trajectory, as percentage).
save_results(dataset_name, method_name, ...)	Save configuration, predictions, and optional evaluation results for a run.
short_time_forecast(truth, prediction, k)	Compute the short-time forecast score (relative L2 over first k steps, as percentage).