6. Add pytest test suite and GitHub Actions CI (Linux, macOS, Windows)

.github/workflows/test-modelmaker.yml

@@ -0,0 +1,69 @@
+name: Tests
+
+on:
+  push:
+    branches: [platypus_dev_1.3, main]
+    paths:
+      - 'tinyml-modelmaker/**'
+      - '.github/workflows/test-modelmaker.yml'
+  pull_request:
+    branches: [platypus_dev_1.3, main]
+    paths:
+      - 'tinyml-modelmaker/**'
+  workflow_dispatch:  # manual trigger
+
+jobs:
+  test:
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, windows-latest, macos-latest]
+    runs-on: ${{ matrix.os }}
+    name: Tests (${{ matrix.os }})
+    # Windows support is experimental — don't block the overall run
+    continue-on-error: ${{ matrix.os == 'windows-latest' }}
+
+    defaults:
+      run:
+        shell: bash
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Python 3.10
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.10'
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip setuptools wheel
+          pip install -e tinyml-modelzoo
+          pip install -e tinyml-tinyverse
+          pip install -e "tinyml-modeloptimization/torchmodelopt"
+          pip install -e tinyml-modelmaker --no-deps
+          pip install defusedxml==0.7.1 "numpy==2.2.6" "PyYAML==6.0.3" "tqdm==4.67.1" "requests==2.32.5" "torch==2.7.1"
+          pip install pytest
+
+      - name: Install ti_mcu_nnc (Linux only)
+        if: runner.os == 'Linux'
+        run: |
+          pip install "ti_mcu_nnc @ https://software-dl.ti.com/mctools/esd/tvm/mcu/ti_mcu_nnc-2.1.1-cp310-cp310-linux_x86_64.whl" || true
+
+      - name: Install ti_mcu_nnc (Windows only)
+        if: runner.os == 'Windows'
+        run: |
+          pip install "ti_mcu_nnc @ https://software-dl.ti.com/mctools/esd/tvm/mcu/ti_mcu_nnc-2.1.1-cp310-cp310-win_amd64.whl" || true
+
+      - name: Tier 1 — Component Tests
+        working-directory: tinyml-modelmaker
+        run: python -m pytest tests/test_model_registry.py tests/test_config_validation.py -v --tb=short
+
+      - name: Tier 3 — Cross-Device Validation
+        working-directory: tinyml-modelmaker
+        run: python -m pytest tests/test_cross_device.py -v --tb=short
+
+      - name: Tier 2 — Pipeline Smoke Tests
+        working-directory: tinyml-modelmaker
+        run: python -m pytest tests/test_pipeline_smoke.py -v --tb=short

tinyml-modelmaker/pyproject.toml

@@ -41,3 +41,10 @@ exclude = [ "tests*",]
 
 [tool.setuptools.dynamic.version]
 attr = "tinyml_modelmaker.__version__"
+
+[tool.pytest.ini_options]
+markers = [
+    "component: Tier 1 component tests",
+    "smoke: Tier 2 pipeline smoke tests",
+    "device: Tier 3 cross-device validation tests",
+]

tinyml-modelmaker/tests/conftest.py

@@ -0,0 +1,181 @@
+"""Shared fixtures for tinyml-modelmaker tests.
+
+Provides:
+- pytest markers for component and smoke test tiers
+- Synthetic dataset generators for pipeline smoke tests
+"""
+
+import os
+import csv
+import random
+import pytest
+from pathlib import Path
+
+
+# ---------------------------------------------------------------------------
+# Custom markers
+# ---------------------------------------------------------------------------
+
+def pytest_configure(config):
+    """Register custom markers for test tiers."""
+    config.addinivalue_line("markers", "component: Tier 1 component tests (fast, no training)")
+    config.addinivalue_line("markers", "smoke: Tier 2 pipeline smoke tests (1-epoch, synthetic data)")
+
+
+# ---------------------------------------------------------------------------
+# Synthetic dataset generators
+# ---------------------------------------------------------------------------
+
+@pytest.fixture
+def synthetic_classification_data(tmp_path):
+    """Create a minimal 3-class classification dataset.
+
+    Structure:
+        tmp_path/
+            classes/
+                class_a/  (30 CSV files)
+                class_b/  (30 CSV files)
+                class_c/  (30 CSV files)
+            annotations/
+                file_list.txt  (all files)
+    Each CSV has 256 rows (samples) × 1 column (single-channel sensor data).
+    """
+    random.seed(42)
+    classes_dir = tmp_path / "classes"
+    all_files = []
+
+    for label in ["class_a", "class_b", "class_c"]:
+        class_dir = classes_dir / label
+        class_dir.mkdir(parents=True)
+        for i in range(30):
+            filename = f"{label}_{i:03d}.csv"
+            filepath = class_dir / filename
+            with open(filepath, "w", newline="") as f:
+                writer = csv.writer(f)
+                for _ in range(256):
+                    writer.writerow([random.gauss(0, 1)])
+            all_files.append(f"classes/{label}/{filename}")
+
+    # Create annotations directory with file list
+    ann_dir = tmp_path / "annotations"
+    ann_dir.mkdir(parents=True)
+    (ann_dir / "file_list.txt").write_text("\n".join(all_files))
+
+    return str(tmp_path)
+
+
+@pytest.fixture
+def synthetic_regression_data(tmp_path):
+    """Create a minimal regression dataset.
+
+    Structure:
+        tmp_path/
+            files/  (60 CSV files — flat layout for regression)
+            annotations/
+                instances_train_list.txt
+                instances_val_list.txt
+    Each CSV has 256 rows × 2 columns (input feature, target value).
+    """
+    random.seed(42)
+    files_dir = tmp_path / "files"
+    files_dir.mkdir(parents=True)
+    all_files = []
+
+    for i in range(60):
+        filename = f"sample_{i:03d}.csv"
+        filepath = files_dir / filename
+        with open(filepath, "w", newline="") as f:
+            writer = csv.writer(f)
+            for _ in range(256):
+                x = random.gauss(0, 1)
+                y = 2.0 * x + random.gauss(0, 0.1)
+                writer.writerow([x, y])
+        all_files.append(f"files/{filename}")
+
+    # Create annotations with train/val split (48 train, 12 val)
+    # File names must match GenericTSDataset glob patterns: *train*_list.txt, *val*_list.txt
+    ann_dir = tmp_path / "annotations"
+    ann_dir.mkdir(parents=True)
+    (ann_dir / "instances_train_list.txt").write_text("\n".join(all_files[:48]))
+    (ann_dir / "instances_val_list.txt").write_text("\n".join(all_files[48:]))
+
+    return str(tmp_path)
+
+
+@pytest.fixture
+def synthetic_anomaly_data(tmp_path):
+    """Create a minimal anomaly detection dataset.
+
+    Structure:
+        tmp_path/
+            classes/
+                normal/  (50 CSV files)
+                anomaly/ (10 CSV files)
+    """
+    random.seed(42)
+    classes_dir = tmp_path / "classes"
+
+    for label, count, mu in [("normal", 50, 0.0), ("anomaly", 10, 5.0)]:
+        class_dir = classes_dir / label
+        class_dir.mkdir(parents=True)
+        for i in range(count):
+            filepath = class_dir / f"{label}_{i:03d}.csv"
+            with open(filepath, "w", newline="") as f:
+                writer = csv.writer(f)
+                for _ in range(256):
+                    writer.writerow([random.gauss(mu, 1)])
+
+    return str(tmp_path)
+
+
+@pytest.fixture
+def synthetic_forecasting_data(tmp_path):
+    """Create a minimal forecasting dataset.
+
+    Structure:
+        tmp_path/
+            files/  (60 CSV files — flat layout for forecasting)
+            annotations/
+                instances_train_list.txt
+                instances_val_list.txt
+    Each CSV has 64 rows × 1 column (univariate time series).
+    """
+    random.seed(42)
+    files_dir = tmp_path / "files"
+    files_dir.mkdir(parents=True)
+    all_files = []
+
+    for i in range(60):
+        filename = f"series_{i:03d}.csv"
+        filepath = files_dir / filename
+        with open(filepath, "w", newline="") as f:
+            writer = csv.writer(f)
+            val = random.gauss(0, 1)
+            for _ in range(64):
+                val += random.gauss(0, 0.3)
+                writer.writerow([val])
+        all_files.append(f"files/{filename}")
+
+    # Create annotations with train/val split (48 train, 12 val)
+    # File names must match GenericTSDataset glob patterns: *train*_list.txt, *val*_list.txt
+    ann_dir = tmp_path / "annotations"
+    ann_dir.mkdir(parents=True)
+    (ann_dir / "instances_train_list.txt").write_text("\n".join(all_files[:48]))
+    (ann_dir / "instances_val_list.txt").write_text("\n".join(all_files[48:]))
+
+    return str(tmp_path)
+
+
+@pytest.fixture
+def modelzoo_examples_dir():
+    """Return the path to the modelzoo examples directory.
+
+    Returns None if not found (not all environments will have modelzoo).
+    """
+    this_dir = Path(__file__).resolve().parent
+    modelmaker_dir = this_dir.parent
+    tensorlab_dir = modelmaker_dir.parent
+    examples_dir = tensorlab_dir / "tinyml-modelzoo" / "examples"
+    if examples_dir.is_dir():
+        return examples_dir
+    return None

tinyml-modelmaker/tests/test_config_dict.py

@@ -0,0 +1,46 @@
+"""Tests for ConfigDict."""
+
+import os
+import tempfile
+
+import pytest
+import yaml
+
+from tinyml_modelmaker.utils.config_dict import ConfigDict
+
+
+class TestConfigDict:
+    def test_from_dict(self):
+        d = {"a": 1, "b": {"c": 2}}
+        cfg = ConfigDict(d)
+        assert cfg.a == 1
+        assert cfg.b.c == 2
+
+    def test_from_yaml(self, tmp_path):
+        data = {"x": 10, "nested": {"y": 20}}
+        yaml_file = tmp_path / "test.yaml"
+        yaml_file.write_text(yaml.dump(data))
+        cfg = ConfigDict(str(yaml_file))
+        assert cfg.x == 10
+        assert cfg.nested.y == 20
+
+    def test_invalid_extension_raises(self, tmp_path):
+        txt_file = tmp_path / "test.txt"
+        txt_file.write_text("hello")
+        with pytest.raises(ValueError, match="unrecognized file type"):
+            ConfigDict(str(txt_file))
+
+    def test_invalid_input_raises(self):
+        with pytest.raises(TypeError, match="got invalid input"):
+            ConfigDict(12345)
+
+    def test_update(self):
+        cfg = ConfigDict({"a": 1, "b": 2})
+        cfg.update({"b": 3, "c": 4})
+        assert cfg.b == 3
+        assert cfg.c == 4
+
+    def test_none_input(self):
+        cfg = ConfigDict(None)
+        # Should create an empty config without error
+        assert isinstance(cfg, ConfigDict)