Release pure-Python ALDes 2.0.0

.github/workflows/ci.yml

@@ -0,0 +1,69 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.11"]
+
+    steps:
+      - uses: actions/checkout@v7
+      - uses: actions/setup-python@v6
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: pip
+      - name: Install ALDes and dependencies
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install -e ".[test]"
+      - name: Verify installed commands and dependencies
+        run: |
+          python -m pip check
+          aldes-train --help
+          aldes-paper-subset --help
+      - name: Lint
+        run: |
+          python -m ruff check .
+          python -m ruff format --check .
+      - name: Test
+        env:
+          ALDES_DEVICE: cpu
+          ALDES_EVAL_WORKERS: "2"
+        run: python -m pytest -q
+
+  package:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: actions/setup-python@v6
+        with:
+          python-version: "3.11"
+          cache: pip
+      - run: python -m pip install --upgrade pip build twine
+      - run: python -m build
+      - run: python -m twine check dist/*
+      - run: python -m venv /tmp/aldes-wheel
+      - run: /tmp/aldes-wheel/bin/python -m pip install dist/*.whl
+      - name: Validate the installed wheel outside the repository
+        working-directory: /tmp
+        run: |
+          /tmp/aldes-wheel/bin/python -m pip check
+          /tmp/aldes-wheel/bin/aldes-train --help
+          /tmp/aldes-wheel/bin/aldes-paper-subset --help
+          /tmp/aldes-wheel/bin/python - <<'PY'
+          import importlib.metadata
+          from run_paper_subset import DEFAULT_REFERENCE_DIR, _reference_result
+
+          assert importlib.metadata.version("aldes") == "2.0.0"
+          assert DEFAULT_REFERENCE_DIR.is_dir()
+          assert _reference_result(DEFAULT_REFERENCE_DIR, 1).size == 30
+          PY

.github/workflows/release.yml

@@ -0,0 +1,26 @@
+name: release assets
+
+on:
+  release:
+    types: [published]
+
+permissions:
+  contents: write
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          ref: ${{ github.event.release.tag_name }}
+      - uses: actions/setup-python@v6
+        with:
+          python-version: "3.11"
+      - run: python -m pip install --upgrade pip build twine
+      - run: python -m build
+      - run: python -m twine check dist/*
+      - name: Attach distributions to the GitHub release
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: gh release upload "${{ github.event.release.tag_name }}" dist/* --clobber

.gitignore

@@ -0,0 +1,15 @@
+.DS_Store
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.ruff_cache/
+.venv/
+build/
+dist/
+*.egg-info/
+logs/
+experiments/
+ela/
+*.pt
+*.pth
+*.ckpt

CHANGELOG.md

@@ -0,0 +1,43 @@
+# Changelog
+
+All notable changes to ALDes are documented here.
+
+## 2.0.0 - 2026-07-20
+
+### Added
+
+- A pure-Python execution path backed by AutoOptLib 1.3.0, with no MATLAB or
+  MATLAB Engine runtime dependency.
+- Automatic PyTorch accelerator selection for NVIDIA CUDA, AMD ROCm, Apple
+  MPS, and CPU, with environment-variable overrides.
+- Deterministic CPU multiprocessing for generated-algorithm and objective
+  evaluation while neural-network work remains on the selected PyTorch device.
+- Explicit independent single-problem and feature-conditioned continual-design
+  modes. Independent design is the default and does not extract problem
+  features; continual design uses landscape features and EWC.
+- A constrained generator and matching executor grammar: fork can follow
+  choose, each branch contains one search operation except that crossover may
+  be followed by mutation, and branches merge into one shared update.
+- Command-line entry points for training and the time-bounded paper subset,
+  portable packaged reference results, tests, and cross-version CI.
+
+### Changed
+
+- Aligned the PBO training protocol with the paper: 100 PPO epochs, 16 sampled
+  algorithms per epoch, five PPO updates, 5,000 training evaluations, and an
+  optional 30-run/50,000-evaluation final test.
+- Replaced the historical Python-to-MATLAB bridge with the public Python
+  AutoOptLib ALDes backend and its 32-token vocabulary.
+- Made one PBO problem and one seed the safe default command-line scope.
+- Moved historical plotting results from the removed MATLAB tree into
+  `draw/datas/reference_results` without changing their contents.
+
+### Removed
+
+- Bundled MATLAB source, MATLAB-specific bridge scripts, stale bytecode,
+  obsolete TorchText data loaders, and generated experiment artifacts.
+
+## 1.0.0
+
+- Original research release combining Python model training with MATLAB
+  algorithm execution.

CITATION.cff

@@ -0,0 +1,40 @@
+cff-version: 1.2.0
+message: "If you use ALDes, please cite the paper below."
+title: "ALDes: Automated Metaheuristic Algorithm Design with Autoregressive Learning"
+type: software
+version: 2.0.0
+date-released: 2026-07-20
+repository-code: "https://github.com/auto4opt/ALDes"
+license: Apache-2.0
+authors:
+  - family-names: Zhao
+    given-names: Qi
+  - family-names: Liu
+    given-names: Tengfei
+  - family-names: Yan
+    given-names: Bai
+  - family-names: Duan
+    given-names: Qiqi
+  - family-names: Yang
+    given-names: Jian
+  - family-names: Shi
+    given-names: Yuhui
+preferred-citation:
+  type: article
+  title: "Automated Metaheuristic Algorithm Design with Autoregressive Learning"
+  year: 2024
+  doi: "10.1109/TEVC.2024.3464677"
+  journal: "IEEE Transactions on Evolutionary Computation"
+  authors:
+    - family-names: Zhao
+      given-names: Qi
+    - family-names: Liu
+      given-names: Tengfei
+    - family-names: Yan
+      given-names: Bai
+    - family-names: Duan
+      given-names: Qiqi
+    - family-names: Yang
+      given-names: Jian
+    - family-names: Shi
+      given-names: Yuhui

EWC.py

@@ -1,60 +1,40 @@
-from copy import deepcopy
+"""Diagonal-Fisher elastic weight consolidation for continual ALDes."""
+
+from __future__ import annotations
 
 import torch
 from torch import nn
-from torch.nn import functional as F
-from torch.autograd import Variable
-import torch.utils.data
-
-
-def variable(t: torch.Tensor, use_cuda=True, **kwargs):
-    if torch.cuda.is_available() and use_cuda:
-        t = t.cuda()
-    return Variable(t, **kwargs)
 
 
-class EWC(object):
+class EWC:
     def __init__(self, model: nn.Module):
-
-        self.model = model
-
-        self.params = {n: p for n, p in self.model.named_parameters() if p.requires_grad}
-        self._means = {}
-        self._precision_matrices = None #self._diag_fisher()
-
-        for n, p in deepcopy(self.params).items():
-            self._means[n] = variable(p.data)
-
-    def _diag_fisher(self):
-        precision_matrices = {}
-        for n, p in deepcopy(self.params).items():
-            p.data.zero_()
-            precision_matrices[n] = variable(p.data)
-
-        #self.model.eval()
-        for n, p in self.model.named_parameters():
-            if p.grad != None:
-                precision_matrices[n].data += p.grad.data ** 2
-        precision_matrices = {n: p for n, p in precision_matrices.items()}
-        return precision_matrices
-
-    def update_diag_fisher(self,model):
-        precision_matrices = {}
-        for n, p in deepcopy(self.params).items():
-            p.data.zero_()
-            precision_matrices[n] = variable(p.data)
-        for n, p in model.named_parameters():
-            if p.grad != None:
-                precision_matrices[n].data += p.grad.data ** 2
-        precision_matrices = {n: p for n, p in precision_matrices.items()}
-        if self._precision_matrices is None:
-            self._precision_matrices = precision_matrices
-        else:
-            for key in precision_matrices:
-                self._precision_matrices[key] +=precision_matrices[key]
-    def penalty(self, model: nn.Module):
-        loss = 0
-        for n, p in model.named_parameters():
-            _loss = self._precision_matrices[n] * (p - self._means[n]) ** 2
-            loss += _loss.sum()
-        return loss*100
+        self._means = {
+            name: parameter.detach().clone()
+            for name, parameter in model.named_parameters()
+            if parameter.requires_grad
+        }
+        self._precision_matrices = {
+            name: torch.zeros_like(parameter)
+            for name, parameter in model.named_parameters()
+            if parameter.requires_grad
+        }
+
+    def update_diag_fisher(self, model: nn.Module) -> None:
+        """Accumulate squared policy gradients for one sampled batch."""
+
+        for name, parameter in model.named_parameters():
+            if name in self._precision_matrices and parameter.grad is not None:
+                self._precision_matrices[name] += parameter.grad.detach().square()
+
+    def penalty(self, model: nn.Module) -> torch.Tensor:
+        loss = torch.zeros((), device=next(model.parameters()).device)
+        for name, parameter in model.named_parameters():
+            if name in self._precision_matrices:
+                loss = (
+                    loss
+                    + (
+                        self._precision_matrices[name]
+                        * (parameter - self._means[name]).square()
+                    ).sum()
+                )
+        return loss