v2.6.0

December 2025

dlordinal v2.6.0 includes the following updates:

---

New Features

• CORN loss function implementation by (JacksonBurns) in (#131)
  Introduced the CORN loss function.

• SORD and SLACE loss functions by (victormvy) in (#144)
  Added SORD and SLACE loss functions.

Improvements & Maintenance

• Python 3.14 support by (victormvy) in (#145)
  Updated the package configuration and CI pipelines to officially support Python 3.14, ensuring compatibility with the latest Python releases and future-proofing the codebase.

• Improved deprecation handling for soft labelling losses
  Deprecated losses, such as BetaCrossEntropyLoss, now correctly raise a DeprecationWarning instead of a FutureWarning. Additionally, the test suite has been updated to suppress these warnings for cleaner output during CI.

---

New Contributors

• @JacksonBurns made his first contribution in #131

Full Changelog: v2.5.0...v2.6.0

v2.5.0

What's Changed

• Enhancement/127 emd loss soft labels by @stefanahaas41 in #128
• BUG: Targets shape correction in hybrid dropout by @franberchez in #125
• [BUG] Fix pre-commit action by @victormvy in #136
• [ENH] Added use_logarithm parameter to WKLoss and updated docstring by @victormvy in #135
• [ENH] Add COPOC output layer and tutorial for unimodal predictions by @stefanahaas41 in #137
• Issue #126 hybrid dropout bug by @franberchez in #138
• [ENH] Added HCI dataset by @victormvy in #140
• [ENH] Add support for torch tensors in metrics calculation by @MohammadElSakka in #134
• Metrics update by @MohammadElSakka in #141
• Metrics update by @franberchez in #142
• [REL] Release 2.5.0 by @franberchez in #143

New Contributors

• @MohammadElSakka made their first contribution in #134

Full Changelog: v2.4.4...v2.5.0

v2.4.4

What's Changed

• [MNT] Added Python 3.13 support by @victormvy in #122

Full Changelog: v2.4.3...v2.4.4

v2.4.3

What's Changed

• [BUG] Fixed losses forward parameters order by @franberchez in #120
• [REL] Release 2.4.3 by @franberchez in #121

Full Changelog: v2.4.2...v2.4.3

v2.4.2

What's Changed

• [ENH] Allow get_beta_soft_labels to accept custom params_set by @victormvy in #109
• [DOC] Loss functions doc improvement by @victormvy in #113
• [ENH] WKLoss optimisation and docstring improvement by @victormvy in #114
• [ENH] WKLoss pred_norm removed by @victormvy in #117
• Replace CLM clipping by stable sigmoid when using logit link function by @RafaAyGar in #116
• Release 2.4.2 by @franberchez in #118

Full Changelog: v2.4.1...v2.4.2

v2.4.1

• Added version consistency check script.
• Fixed version string in __init__.py.

v2.4.0

April 2025

dlordinal v2.4.0 includes the following updates:

---

Loss Functions

• Implemented the CDWCE loss function.
• Fixed an issue in WKLoss that prevented it from working when y_true was one-hot encoded.
• Resolved a device inconsistency in the WKLoss.forward method.
• Added squared Earth Mover’s Distance (EMD) loss.
• Extended WKLoss to support both logits and probabilities (previously supported only probabilities).

---

Soft Labelling

• Added examples for geometric soft labels.
• Refactored soft labelling losses:
  These have been reimplemented to support arbitrary loss functions, not just CrossEntropyLoss. Additionally, they have been renamed for clarity:
  • BetaCrossEntropyLoss → BetaLoss
  • BinomialCrossEntropyLoss → BinomialLoss
  • CustomTargetsCrossEntropyLoss → CustomTargetsLoss
  • ExponentialCrossEntropyLoss → ExponentialLoss
  • GeneralTriangularCrossEntropyLoss → GeneralTriangularLoss
  • PoissonCrossEntropyLoss → PoissonLoss
  • TriangularCrossEntropyLoss → TriangularLoss

⚠️ The original soft labelling loss classes are now deprecated and scheduled for removal in v3.0.0.

---

Maintenance

• Added GPU-enabled versions of all test cases.

v2.3.2

What's Changed

• [BUG] AMAE and MMAE correction by @franberchez in #95
• Release v2.3.2 by @franberchez in #96

Full Changelog: v2.3.1...v2.3.2

v2.3.1

What's Changed

• [BUG] Added support for any number of classes in get_binomial_soft_labels() by @victormvy in #93

Full Changelog: https://github.com/ayrna/dlordinal/blob/main/CHANGELOG.md

v2.3.0

What's Changed

• [MNT] Update pre-commit configuration by @victormvy in #89
• [MNT] Modified tutorials github action to run skorch tutorials separately by @victormvy in #90
• Enhancement/87 add geometric loss by @stefanahaas41 in #88
• [MNT] pre-commit workflow by @victormvy in #91
• [MNT] Release v2.3.0 by @franberchez in #92

New Contributors

• @stefanahaas41 made their first contribution in #88

Full Changelog: v2.2.1...v2.3.0