0.3.0: better small-data accuracy, calibration, active learning, serving, sklearn API, and a hardened test suite by codelion · Pull Request #68 · codelion/adaptive-classifier

Modernize packaging and add deployment guide

claude

committed

d2ae81c

Save resolved pooling and tokenizer files with the model

claude

committed

2276951

Add forget, remove_examples and out-of-distribution detection

claude

committed

6af5f5a

CI: run every test, cache models, add timeouts and a weekly run

claude

committed

753ab30

Add scikit-learn compatible SklearnAdaptiveClassifier

claude

committed

62a31dd

Bump to 0.3.0, make CI offline and stable, add robustness and failure-mode tests

claude

committed

47007c0

CI: prefetch models under canonical ids, fail fast on permanent errors

claude

committed

6f08727

Fix small-data accuracy: train the head properly, sharpen prototype scores, ramp the head in

claude

committed

fb8b334

Add confidence calibration and conformal prediction sets

claude

committed

8e50c99

Add active learning (suggest_labels) and drift detection (drift_report)

claude

committed

a725359

Add HTTP serving, a thread-safe classifier, async methods and a Dockerfile

claude

committed

9013ac4

Add uvicorn to the test extra

claude

committed

76c7c1b

Add a few-shot benchmark against SetFit and logistic regression

claude

committed

2132d33

Use a placeholder encoder name in the benchmark harness tests

claude

committed

83a799e

Remove the SetFit comparison from the README and changelog

claude

committed

9fff1f0

Keep the fixed new-class split when loading saves from before 0.2.0

claude

committed

4a60c49

Navigation Menu

Repository navigation

0.3.0: better small-data accuracy, calibration, active learning, serving, sklearn API, and a hardened test suite - #68

0.3.0: better small-data accuracy, calibration, active learning, serving, sklearn API, and a hardened test suite#68
codelion merged 16 commits into
mainfrom
claude/zen-hawking-qj5dfq

Commits on Oct 8, 2026

Modernize packaging and add deployment guide

Save resolved pooling and tokenizer files with the model

Add forget, remove_examples and out-of-distribution detection

CI: run every test, cache models, add timeouts and a weekly run

Add scikit-learn compatible SklearnAdaptiveClassifier

Bump to 0.3.0, make CI offline and stable, add robustness and failure-mode tests

CI: prefetch models under canonical ids, fail fast on permanent errors

Fix small-data accuracy: train the head properly, sharpen prototype scores, ramp the head in

Add confidence calibration and conformal prediction sets

Add active learning (suggest_labels) and drift detection (drift_report)

Add HTTP serving, a thread-safe classifier, async methods and a Dockerfile

Add uvicorn to the test extra

Add a few-shot benchmark against SetFit and logistic regression

Use a placeholder encoder name in the benchmark harness tests

Remove the SetFit comparison from the README and changelog

Keep the fixed new-class split when loading saves from before 0.2.0