From fccffd383261319e40e9e22338d2b1bfda2197d5 Mon Sep 17 00:00:00 2001
From: jjscripts <pardojeromeimportant@gmail.com>
Date: Wed, 3 Jun 2026 16:32:39 +0800
Subject: [PATCH] release: 0.2.0 (transfer entropy, divergence, matplotlib
 backend fix)

Bump 0.1.1 -> 0.2.0 with a CHANGELOG entry covering the transfer-entropy and
divergence modules, the matplotlib Agg-on-import fix (#2), and the new examples.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 CHANGELOG.md           | 25 +++++++++++++++++++++++++
 entroscope/__init__.py |  2 +-
 pyproject.toml         |  2 +-
 3 files changed, 27 insertions(+), 2 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index a44b7f1..c3a2363 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,30 @@
 # Changelog
 
+## 0.2.0 — 2026-06-03
+
+- **Transfer entropy** (`entroscope.transfer`) — the first bivariate measure:
+  directional information flow `TE(X→Y)` (Schreiber 2000) with two estimators,
+  Kraskov KSG k-NN (default, no binning) and a binned histogram cross-check.
+  API: `compute`, `rolling`, `delta`, `plot`. A sample-size guard warns when the
+  data is too thin for the embedding dimension. Correctness is validated against
+  three independent authorities: the bivariate-Gaussian closed form, the
+  Kraskov-2004 analytic mutual information, and a hand-checked isolated embedding.
+- **Divergence** (`entroscope.divergence`) — distribution-vs-distribution
+  measures: `kl` (Kullback-Leibler, directional) and `js` (Jensen-Shannon,
+  symmetric, bounded `[0, 1]` bits), plus `plot`. Inputs are two raw samples,
+  binned over a shared range; KL uses epsilon smoothing to stay finite on empty
+  bins. Useful for data-drift detection (training vs production distributions).
+- **Fixed:** importing a measure no longer forces matplotlib's `Agg` backend
+  (issue #2). The library no longer mutates global matplotlib state on import, so
+  interactive plotting (e.g. `df.plot()` in a notebook) keeps working. Headless,
+  Docker, and CI users who want a guaranteed non-interactive backend should set
+  `MPLBACKEND=Agg` in their environment.
+- **Correlation-stability example** (`examples/correlation_stability.py`) —
+  entropy of a rolling correlation series as a regime-stability gauge, shipped
+  alongside a rolling-std baseline so the comparison stays honest.
+- New worked examples: `examples/information_flow.py` (directional transfer
+  entropy) and `examples/drift.py` (KL/JS distribution drift).
+
 ## 0.1.1 — 2026-06-01
 
 - Added a `LICENSE` file (MIT) and a `CONTRIBUTING.md` guide.
diff --git a/entroscope/__init__.py b/entroscope/__init__.py
index 704afa6..d35d261 100644
--- a/entroscope/__init__.py
+++ b/entroscope/__init__.py
@@ -13,7 +13,7 @@
 )
 from .utils import plot
 
-__version__ = "0.1.1"
+__version__ = "0.2.0"
 __all__ = [
     "shannon",
     "permutation",
diff --git a/pyproject.toml b/pyproject.toml
index 254eb17..ea733c2 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "entroscope"
-version = "0.1.1"
+version = "0.2.0"
 description = "The definitive entropy toolkit for time series data"
 readme = "README.md"
 requires-python = ">=3.9"