src.tests.torf_losses_test

Classes

TorfLossesTest()

class src.tests.torf_losses_test.TorfLossesTest

Author:: Alberto M. Esmoris Pena

Test that TORF masked losses produce the same results as their non-TORF counterparts on fully valid data, and correctly exclude masked positions.

__init__()

Basic configuration for any VL3D test.

Parameters:: name (str) – Test name

run()

Run all TORF loss tests.

Returns:: True if all tests pass, False otherwise.
Return type:: bool

static test_torf_cce(eps): Plain categorical: TORF vs Keras on valid data.

static test_torf_cwcce(eps): CW categorical: TORF vs non-TORF on valid data.

static test_torf_fcce(eps): Focal categorical: TORF vs numpy on valid data.

static test_torf_cwfcce(eps): CW+focal categorical: TORF vs non-TORF.

static test_torf_bce(eps): Plain binary: TORF vs Keras on valid data.

static test_torf_cwbce(eps): CW binary: TORF vs non-TORF on valid data.

static test_torf_fbce(eps): Focal binary: TORF vs numpy on valid data.

static test_torf_cwfbce(eps): CW+focal binary: TORF vs non-TORF.

static test_masking_categorical(eps): Verify masked positions are excluded.

static test_masking_binary(eps): Verify masked positions are excluded.

static test_masking_all_categorical(eps): Masking for all 4 categorical variants.

static test_masking_all_binary(eps): Masking for all 4 binary variants.

static test_scattered_mask(eps): Non-contiguous (scattered) mask pattern.

static test_single_valid_position(eps): Single valid position per batch element.

static test_rank3_binary(eps): Binary with rank-3 input (B, K, 1).

static test_rank3_binary_masked(eps): Rank-3 binary (B,K,1) with masking.

static test_different_gamma(eps): Focal with gamma != 2.0 (test 1.5 and 3.0).

static test_two_class_categorical(eps): Categorical with nc=2 (boundary case).

static test_all_invalid(): All-invalid batch returns 0.0, not NaN.