Loss-Guided TM

Models

MultiClassTM(n_clauses, s, dim, n_classes, **opt)	Guided TM for multi-class classification: each sample has exactly one of n_classes classes.
MultiOutputTM(n_clauses, s, dim, n_classes, ...)	Guided TM for multi-label classification: each sample can have any number of n_classes labels.
BinaryTM(n_clauses, s, dim, **opt)	Guided TM for binary classification: one output, each sample is 0 or 1.
RegressionTM(n_clauses, s, dim, **opt)	Guided TM for regression: one real-valued target per sample.

Configuration

TMConfig()	Settings for the guided TM models.
ActFn(*values)
FbSignal(*values)

Activations and losses

ActLoss()	Base class for an output activation paired with a loss.
SoftmaxCE([eps, weights])	Softmax with cross-entropy, for single-label classification.
SigmoidBCE([eps, pos_weights, neg_weights])	Sigmoid with binary cross-entropy per class, for binary and multi-label classification.
MSE()	Identity with squared error, for regression.
MAE()	Identity with absolute error, for regression.
Huber([delta])	Identity with Huber loss, for regression.
ASL([gamma_pos, gamma_neg, clip, eps])	Sigmoid with asymmetric loss, for multi-label classification with many negatives.
SCE([alpha, beta, eps, eps_hat])	Softmax with symmetric cross-entropy (cross-entropy plus reverse cross-entropy), for noisy labels.
Tversky([alpha, beta, eps])	Sigmoid with Tversky loss over the classes of a sample, for multi-label classification.
FocalBCE([alpha, gamma, eps])	Sigmoid with focal binary cross-entropy per class, for imbalanced binary and multi-label classification.
FocalCE([alpha, gamma, eps])	Softmax with focal cross-entropy, for imbalanced single-label classification.