What’s new#

v0.2.1 (March 2026)#

Bug fixes

Fix off-by-one in antropy.higuchi_fd(): the inner loop summed $N_m - 1$ differences per sub-series instead of the correct $N_m$, causing a systematic underestimate of curve length. Verified against the NeuroKit2 reference implementation.
antropy.sample_entropy() now returns np.nan (instead of 0) when no template of length m matches within the tolerance, since the entropy is mathematically undefined in that case.
Fix silent epsilon = 10e-9 (= 1e-8) typo in utils.py; corrected to 1e-9. This constant guards against division-by-zero in the Numba linear regression used by antropy.detrended_fluctuation().
Fix all = [...] → __all__ = [...] in all three source modules, which previously leaked imported names (np, jit, KDTree, …) into the public namespace on from antropy import *.

New features

antropy.hjorth_params() gains an optional sf parameter. When provided, mobility is returned in Hz instead of samples⁻¹ (multiplied by sf); complexity is unaffected.

Validation

User-facing assert statements replaced with proper ValueError / TypeError raises throughout (asserts are silently stripped by python -O): perm_entropy(delay=0), spectral_entropy(method=…), app_entropy/sample_entropy(tolerance=…), and lziv_complexity(sequence=…, normalize=…).

Docs

Docstring formula for antropy.higuchi_fd() corrected to match the fix above ($\sum_{j=1}^{N_m}$ not $\sum_{j=1}^{N_m-1}$).
Added notes to antropy.lziv_complexity() (float arrays are truncated to uint32, not discretised — binarise continuous signals first), antropy.num_zerocross() (signbit(0) == False behaviour), antropy.spectral_entropy() (DC component included), antropy.detrended_fluctuation() ($\alpha \approx 1$ is ambiguous), and antropy.hjorth_params() (mobility units).

Build & CI

Drop Python 3.9 (EOL), add Python 3.13 support. Minimum is now Python 3.10.
Switch from pip to uv in all GitHub Actions workflows.
Add explicit minimum versions for core dependencies: numpy>=1.22.4, scipy>=1.8.0, scikit-learn>=1.2.0.
Migrate [project.optional-dependencies] to PEP 735 [dependency-groups].
Bump setuptools>=80.0.
Split CI into three jobs: test-core (3 platforms × 4 Python versions), test-dependency-combinations (4 dep combos from minimum to latest), and coverage.
Fix test-dependency-combinations job: separate antropy install (--no-deps) from test-dependency install so that pytest’s own dependencies (e.g. pluggy) are always resolved.
Fix Codecov upload to use ${{ secrets.CODECOV_TOKEN }} instead of a hardcoded token.
Switch Ruff workflow from astral-sh/ruff-action@v1 to uvx ruff via astral-sh/setup-uv@v7.
Extend Ruff rules: add W (pycodestyle warnings) and NPY (NumPy rules).

Tests

Increase test coverage from ~54 % to 100 %.
Add tests/test_utils.py covering all branches of the _embed helper (1-D and 2-D paths, all error conditions).
Add edge-case tests: sample_entropy returning inf (m-length matches exist but no (m+1)-length matches); detrended_fluctuation returning NaN for a constant signal; spectral_entropy raising on an invalid method string.
Set NUMBA_DISABLE_JIT=1 in the coverage CI job so coverage.py can instrument Numba JIT function bodies; other CI jobs still exercise real compiled code.

Bug fixes

Fix antropy.higuchi_fd() returning a ValueError (math domain error) on constant or integer-typed input arrays: log(0) is now guarded to return -inf, matching Numba’s IEEE 754 behaviour.

Docs

Switch documentation theme from sphinx_bootstrap_theme to pydata-sphinx-theme (dark/light toggle, GitHub icon, improved layout).
Fix three broken intersphinx URLs: NumPy, SciPy, and MNE-Python.
Add sphinx.ext.mathjax for LaTeX math rendering in docstrings.
Add contributing.rst guide.
Fix stale :py:func: cross-references in changelog entries v0.1.1–v0.1.3 (entropy.XXX → antropy.XXX).
Fix typo in v0.1.6 changelog (“Fox for KDTree” → “Fix for KDTree”).
Modernize README.rst and docs/index.rst: add PyPI, conda-forge, downloads, and Ruff badges; add uv installation instructions; fix broken links.

Simplify Katz FD implementation. raphaelvallat/antropy#30
Add tolerance parameter to the approximate and sample entropy. raphaelvallat/antropy#32
Fix for scikit-learn ≥ 1.3 in approximate and sample entropy. raphaelvallat/antropy#36

This version requires numba >= 0.57.

Allow readonly arrays in numba jit signature. raphaelvallat/antropy#23
Improved sample entropy kernel. raphaelvallat/antropy#25
Fix for KDTree.valid_metrics which is method since sklearn 1.3. raphaelvallat/antropy#30

antropy.perm_entropy() will now return the average entropy across all delays if a list or range of delays is provided.
Handle the limit of p = 0 in functions that evaluate the product p * log2(p), to give 0 instead of nan (see PR3).
antropy.detrended_fluctuation() will now return alpha = 0 when the correlation coefficient of the fluctuations of an input signal is 0 (see PR21).

Important

The package has now been renamed AntroPy (previously EntroPy)!

Faster implementation of antropy.lziv_complexity() (see PR1). Among other improvements, strings are now mapped to UTF-8 integer representations.

Added the antropy.num_zerocross() function to calculate the (normalized) number of zero-crossings on N-D data.
Added the antropy.hjorth_params() function to calculate the mobility and complexity Hjorth parameters on N-D data.
Add support for N-D data in antropy.spectral_entropy(), antropy.petrosian_fd() and antropy.katz_fd().
Use the stochastic package to generate stochastic time-series.

antropy.lziv_complexity() now works with non-binary sequence (e.g. “12345” or “Hello World!”)
The average fluctuations in antropy.detrended_fluctuation() is now calculated using the root mean square instead of a simple arithmetic. For more details, please refer to this GitHub issue.
Updated flake8

Initial release.