TabICL 2.0.1

What's Changed

• Fix two CUDA stream race conditions in AsyncCopyManager for cpu/disk offloading (Issue #34)

Full Changelog: v2.0.0...v2.0.1

TabICL 2.0.0

Release the code of TabICLv2

Full Changelog: v0.1.4...v2.0.0

TabICL 0.1.4

• Introduce sklearn version-agnostic utilities

• Update Python version range to support up to 3.13

Full Changelog: v0.1.3...v0.1.4

TabICL 0.1.3

• fix (InferenceManager): Add no_grad context for CPU inference
• fix (TabICLClassifier): Correct thread count calculation

Full Changelog: v0.1.2...v0.1.3

v0.1.2

What's Changed

• Fix bug when the number of classes exceeds 100 (Issue #8)
• Fix recursion error in the _latin_squares method of FeatureShuffler (Issue #9)
• docs: add alternate installation option by @TheFloatingString in #10

Full Changelog: v0.1.1...v0.1.2

TabICL 0.1.1

What's Changed

Add a better-performing checkpoint

Full Changelog: v0.1.0...v0.1.1

TabICL 0.1.0

What's Changed

Open-source pretraining code

Full Changelog: v0.0.6...v0.1.0

TabICL 0.0.6

What's Changed

• fix (TabICLClassifier): convert output tensor to float before moving to numpy (df1e504)

When AMP is enabled, PyTorch tensors may be in float16. Explicitly casting to float32 ensures stability and prevents errors.

Full Changelog: v0.0.5...v0.0.6

TabICL 0.0.5

What's Changed

• Fix: Normalize probabilities of TabLCLClassifier's predict_proba to sum to 1 (44c2d27)

Full Changelog: v0.0.4...v0.0.5

TabICL 0.0.4

What's Changed

• Add RTDL quantile transform by @dholzmueller in #5

Full Changelog: v0.0.3...v0.0.4