Crate tsai_analysis 

tsai_analysis

Analysis utilities for tsai-rs: confusion matrix, top losses, permutation importance.

This crate provides tools for analyzing model performance:

• Confusion matrix computation and visualization
• Top losses identification
• Feature and step importance via permutation
• Calibration analysis (ECE, MCE, temperature scaling)
• Classification report (per-class precision, recall, F1)
• Time series feature extraction (tsfresh-style)

Re-exports

pub use features::extract_features;

pub use features::extract_multivariate_features;

pub use features::FeatureExtractor;

pub use features::FeatureSet;

Modules

features

Time series feature extraction similar to tsfresh.

Structs

CalibrationResult

Calibration analysis results.

ClassMetrics

Per-class classification metrics.

ClassificationReport

Classification report with per-class and aggregate metrics.

ConfusionMatrix

Confusion matrix for classification evaluation.

PermutationImportance

Importance score for a feature or time step.

TopLoss

A sample with its loss value.

Functions

calibration_from_probs

Compute calibration from probability matrix.

classification_report

Compute a classification report from predictions and targets.

compute_calibration

Compute calibration metrics for classification predictions.

confusion_matrix

Compute confusion matrix from predictions and targets.

feature_importance

Compute feature (variable) importance via permutation.

find_optimal_temperature

Find optimal temperature for calibration.

step_importance

Compute time step importance via permutation.

temperature_scale

Temperature scaling for calibration.

top_losses

Get the top losses (samples with highest loss).