Module aggregate 

Temperature-scaled softmax and the classification/regression ensemble-combination paths from TabFMClassifier._process_logits / TabFMRegressor._combine_predictions.

Enums

ClassAggMode

Functions

aggregate_classification

_process_logits: logits_all is [n_estimators][n_classes], already un-shifted back to original class order. Returns final [n_classes] probabilities.

average_scaled_predictions

_combine_predictions, NNLS-off (default) path: average the still-scaled per-member predictions first; the caller inverse-transforms the single averaged result afterward.

softmax_temperature

TabFMClassifier.softmax (temperature divides logits before the max-subtracted softmax).

unshift_logits

Undoes a member’s class-label shift: that member was fed (y_true + shift) % n_classes as training labels, so its output logit at position (c + shift) % n_classes is the prediction for original class c.

weighted_unscaled_predictions

_combine_predictions, NNLS-on path: each member’s prediction has already been inverse-transformed by the caller; combine via the fitted weights.