Release v0.3.0: deprecate AdaptiveModeAdamW/SGD wrappers

CHANGELOG.md

@@ -0,0 +1,95 @@
+# Changelog
+
+All notable changes to this project are documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.3.0] - 2026-05-11
+
+### Deprecated
+- `AdaptiveModeAdamW` and `AdaptiveModeSGD` are deprecated and will be removed
+  in 0.4.0. Use `AEES`, which wraps any `torch.optim.Optimizer`. Importing or
+  instantiating either class now emits a `DeprecationWarning`.
+
+## [0.2.2] - 2026-05-10
+
+Patch release with packaging cleanup and security hardening.
+
+### Added
+- `MANIFEST.in` to refine PyPI source distribution contents.
+- Runnable examples included in the source distribution (wheel kept minimal).
+
+### Changed
+- Clarified README documentation around examples and repository layout.
+- Updated test configuration.
+- Pinned third-party GitHub Actions to commit SHAs to reduce CI supply-chain risk.
+
+### Fixed
+- Replaced an unsafe `assert` state check in `Scheduler.step_end` with an
+  explicit runtime check.
+
+## [0.2.1] - 2026-05-07
+
+Metadata-only patch.
+
+### Added
+- `Changelog` URL under `[project.urls]` so PyPI surfaces a link to the
+  changelog in the project sidebar.
+
+## [0.2.0] - 2026-05-07
+
+Breaking-change release that removes the `trend_phase` context mode.
+
+### Removed
+- `pulseopt.TREND_PHASE_CONTEXT_BUCKETS`.
+- `context_mode="trend_phase"`.
+- `total_training_steps` kwarg on `AEES` and `StructuredEpisodeManager` (was
+  only used to compute the `trend_phase` boundary).
+- `StructuredSelection.context_bucket_id`, `.context_bucket_name`,
+  `.context_phase`, `.context_trend`.
+- Episode log keys `context_bucket_ids`, `context_bucket_names`,
+  `context_phases`, `context_trends`.
+
+### Changed
+- `StructuredSelection` now exposes a single `context_bucket: str | None`
+  field; episode logs expose a single `context_buckets` list (only present
+  when `context_mode != "none"`).
+
+### Migration
+- If you used `context_mode="trend_phase"`, switch to `"trend"` (or `"none"`).
+- If you read any of the removed `StructuredSelection` fields or log keys,
+  read `context_bucket` / `context_buckets` instead.
+
+## [0.1.6rc1] - 2026-05-06
+
+### Added
+- CI/CD workflows and ruff lint/format configuration.
+- Release smoke test for the CI/CD pipeline.
+
+## [0.1.5] - 2026-05-06
+
+### Removed
+- Files irrelevant to the PyPI package that remained from thesis experiments.
+
+### Changed
+- Default `reward_instability_lambda` set to `0.0` in `reward.py`.
+
+## [0.1.4] - 2026-05-05
+
+### Fixed
+- Changed repo-relative paths to absolute.
+
+### Changed
+- Set reward instability constant to `0.0`.
+
+[Unreleased]: https://github.com/davidkfoss/pulseopt/compare/v0.3.0...HEAD
+[0.3.0]: https://github.com/davidkfoss/pulseopt/compare/v0.2.2...v0.3.0
+[0.2.2]: https://github.com/davidkfoss/pulseopt/compare/v0.2.1...v0.2.2
+[0.2.1]: https://github.com/davidkfoss/pulseopt/compare/v0.2.0...v0.2.1
+[0.2.0]: https://github.com/davidkfoss/pulseopt/compare/v0.1.6rc1...v0.2.0
+[0.1.6rc1]: https://github.com/davidkfoss/pulseopt/compare/v0.1.5...v0.1.6rc1
+[0.1.5]: https://github.com/davidkfoss/pulseopt/compare/v0.1.4...v0.1.5
+[0.1.4]: https://github.com/davidkfoss/pulseopt/releases/tag/v0.1.4

pyproject.toml

@@ -45,7 +45,7 @@ dev = [
 Homepage = "https://github.com/davidkfoss/pulseopt"
 Repository = "https://github.com/davidkfoss/pulseopt"
 Issues = "https://github.com/davidkfoss/pulseopt/issues"
-Changelog = "https://github.com/davidkfoss/pulseopt/releases"
+Changelog = "https://github.com/davidkfoss/pulseopt/blob/main/CHANGELOG.md"
 
 [tool.setuptools]
 include-package-data = true

src/pulseopt/__init__.py

@@ -1,6 +1,6 @@
 """Adaptive Episodic Exploration Scheduling package."""
 
-__version__ = "0.2.2"
+__version__ = "0.3.0"
 
 from pulseopt.controller import (
     TREND_CONTEXT_BUCKETS,

src/pulseopt/optimizer.py

@@ -1,7 +1,14 @@
-"""Scheduler-compatible optimizer wrappers with AEES mode scaling."""
+"""Scheduler-compatible optimizer wrappers with AEES mode scaling.
+
+.. deprecated:: 0.3.0
+    ``AdaptiveModeAdamW`` and ``AdaptiveModeSGD`` are deprecated and will be
+    removed in 0.4.0. Use :class:`pulseopt.AEES`, which wraps any
+    ``torch.optim.Optimizer``.
+"""
 
 from __future__ import annotations
 
+import warnings
 from collections.abc import Iterable
 from copy import deepcopy
 from typing import Any
@@ -11,9 +18,18 @@
 
 from pulseopt.types import CandidateConfig
 
+_DEPRECATION_MESSAGE = (
+    "{cls} is deprecated and will be removed in pulseopt 0.4.0. "
+    "Use pulseopt.AEES, which wraps any torch.optim.Optimizer."
+)
+
 
 class AdaptiveModeAdamW(torch.optim.AdamW):
-    """AdamW optimizer that applies AEES multipliers on top of scheduled bases."""
+    """AdamW optimizer that applies AEES multipliers on top of scheduled bases.
+
+    .. deprecated:: 0.3.0
+        Use :class:`pulseopt.AEES` instead. Will be removed in 0.4.0.
+    """
 
     def __init__(
         self,
@@ -26,6 +42,11 @@ def __init__(
         mode: CandidateConfig | None = None,
         noise_seed: int | None = None,
     ) -> None:
+        warnings.warn(
+            _DEPRECATION_MESSAGE.format(cls="AdaptiveModeAdamW"),
+            DeprecationWarning,
+            stacklevel=2,
+        )
         if not torch.isfinite(torch.tensor(float(lr))):
             raise ValueError("lr must be finite.")
         if lr <= 0.0:
@@ -269,7 +290,11 @@ def _restore_base_list(
 
 
 class AdaptiveModeSGD(torch.optim.SGD):
-    """SGD optimizer that applies AEES multipliers on top of scheduled bases."""
+    """SGD optimizer that applies AEES multipliers on top of scheduled bases.
+
+    .. deprecated:: 0.3.0
+        Use :class:`pulseopt.AEES` instead. Will be removed in 0.4.0.
+    """
 
     def __init__(
         self,
@@ -282,6 +307,11 @@ def __init__(
         mode: CandidateConfig | None = None,
         noise_seed: int | None = None,
     ) -> None:
+        warnings.warn(
+            _DEPRECATION_MESSAGE.format(cls="AdaptiveModeSGD"),
+            DeprecationWarning,
+            stacklevel=2,
+        )
         if not torch.isfinite(torch.tensor(float(lr))):
             raise ValueError("lr must be finite.")
         if lr <= 0.0:

tests/test_optimizer_wrapper.py

@@ -2,11 +2,14 @@
 
 from __future__ import annotations
 
+import pytest
 import torch
 
 from pulseopt.optimizer import AdaptiveModeAdamW
 from pulseopt.types import CandidateConfig
 
+pytestmark = pytest.mark.filterwarnings("ignore:AdaptiveModeAdamW is deprecated:DeprecationWarning")
+
 
 def build_model(weight: float = 1.0) -> torch.nn.Linear:
     """Create a tiny deterministic linear model."""