Confromity Update

Author: HakimBenkirane

.claude/projects/-Users-hakimbenkirane-Documents-Softwares-CustOmics/memory/MEMORY.md

@@ -0,0 +1 @@
+- [customics package refactor complete](project_state.md) — v0.1.0 PyPI-ready, bc_multiomics env, 54 passing tests

.claude/projects/-Users-hakimbenkirane-Documents-Softwares-CustOmics/memory/project_state.md

@@ -0,0 +1,25 @@
+---
+name: customics package refactor complete
+description: CustOmics was refactored from a research repo into a proper PyPI-ready package named customics (v0.1.0)
+type: project
+---
+
+The repo was fully refactored from a research-only codebase into an installable `customics` package (v0.1.0).
+
+**Why:** User wants to publish to PyPI.
+
+**What was done:**
+- New layout: `src/customics/` package with `__init__.py` in every sub-package
+- `pyproject.toml` with `setuptools.build_meta` backend, Python ≥ 3.9
+- Active conda env for this project: `bc_multiomics`
+- Run tests with: `conda run -n bc_multiomics python -m pytest tests/`
+
+**Key bugs fixed:**
+- B1: `list` → `nn.ModuleList` for autoencoders (fixes `state_dict()`)
+- B2: `eval()` → `_ACTIVATIONS` dict in `SurvivalNet`
+- B3: `classification_loss` arg order corrected (`y_pred, y_true`)
+- B4: `FullyConnectedLayer` now honours the passed `norm_layer`
+- B5: Broken `evaluate_latent()` removed; `evaluate()` unified
+- OHE fitted on integer-encoded labels (not strings) in `fit()`
+
+**How to apply:** When working on this repo, install in editable mode in bc_multiomics and run pytest to verify.

.github/workflows/ci.yml

@@ -0,0 +1,33 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.11"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install package and dev dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e ".[dev]"
+
+      - name: Run tests
+        run: pytest --tb=short -q
+
+      - name: Check package imports
+        run: python -c "import customics; print(customics.__version__)"

CHANGELOG.md

@@ -0,0 +1,69 @@
+# Changelog
+
+All notable changes to `customics` are documented here.
+
+Format: [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) — versioning follows [Semantic Versioning](https://semver.org/).
+
+---
+
+## [0.1.0] — 2026-04-21
+
+First PyPI-ready release. The package was refactored from the original research
+codebase with full packaging, typing, testing, and documentation.
+
+### Added
+- `pyproject.toml` — installable via `pip install customics`
+- Proper `src/customics/` package layout with `__init__.py` in every sub-package
+- `__version__ = "0.1.0"` version string
+- `customics` CLI entry point (`python -m customics` or `customics` command)
+- `customics.exceptions` module: `CustOmicsError`, `DataValidationError`,
+  `ModelNotFittedError`, `ConfigurationError`
+- Type hints on all public methods and class signatures
+- `logging` integration throughout (replaces all `print()` calls)
+- Input validation in `CustOMICS.__init__` and `fit()` with descriptive errors
+- `CustOMICS.predict()` — dedicated classification prediction method
+- `CustOMICS.source_names` attribute for explicit source-to-index mapping
+- `CustOMICS._get_central_representation()` — always uses the central VAE for
+  inference regardless of training phase
+- `customics.visualization` module: `plot_loss`, `plot_representation`,
+  `plot_survival_stratification` (extracted from the monolithic model class)
+- `customics.explain.shap` module (replaces `ex_vae/shap_vae.py`)
+- `MultiOmicsDataset.get_samples()` method
+- Complete NumPy-style docstrings with `Parameters`, `Returns`, `Raises`, and
+  `Examples` sections throughout
+- Full test suite: unit tests for losses, metrics, dataset, encoders, model;
+  integration tests for fit → evaluate → save/load
+- GitHub Actions CI workflow (Python 3.9 and 3.11)
+- `CHANGELOG.md` and `CONTRIBUTING.md`
+- Updated `README.md` with installation, quick start, API table, and data-format
+  documentation
+
+### Fixed
+- **[B1]** `self.autoencoders`, `self.lt_encoders`, `self.lt_decoders` were plain
+  Python `list` objects — replaced with `nn.ModuleList` so `state_dict()`,
+  `load_state_dict()`, and `to(device)` work correctly
+- **[B2]** `eval('nn.{}()'.format(activation))` in `SurvivalNet` — replaced with
+  a safe dictionary dispatch (`_ACTIVATIONS` map)
+- **[B3]** `classification_loss` argument order was inverted (`y_true`/`y_pred`
+  swapped relative to call sites) — corrected signature and docstring
+- **[B4]** `FullyConnectedLayer` silently overwrote the passed `norm_layer`
+  argument with `nn.BatchNorm1d` — fixed to honour the caller's choice
+- **[B5]** `evaluate_latent()` called `svc_model.predict_proba()` without
+  arguments — removed the broken method; `evaluate()` now correctly handles both
+  tasks
+
+### Changed
+- Package name standardised to lowercase `customics` throughout
+- All `from src.xxx import` statements updated to `from customics.xxx import`
+- `Decoder` and `ProbabilisticDecoder` now reverse `hidden_dim` internally,
+  mirroring the encoder ordering
+- `fit()` now returns `self` for method chaining
+- `CustOMICS.stratify()` and `CustOMICS.plot_*` are thin wrappers that delegate
+  to `customics.visualization`
+- `src/tools/prepare_dataset.py` and `src/tools/core_utils.py` removed (tightly
+  coupled to internal TCGA paths; not library code)
+- `src/debug/print_layer.py` removed (unused)
+- `src/config/samples.txt` dependency removed from `utils.py`
+
+### Deprecated
+- `evaluate_latent()` — removed (was broken; use `evaluate()` instead)

CONTRIBUTING.md

@@ -0,0 +1,57 @@
+# Contributing to customics
+
+Thank you for your interest in contributing!
+
+---
+
+## Development Setup
+
+```bash
+git clone https://github.com/HakimBenkirane/CustOmics.git
+cd CustOmics
+pip install -e ".[dev]"
+```
+
+This installs the package in editable mode together with `pytest` and `pytest-cov`.
+
+---
+
+## Running Tests
+
+```bash
+pytest                          # run all tests
+pytest tests/unit/              # unit tests only
+pytest tests/integration/       # integration tests only
+pytest --cov=customics          # with coverage report
+```
+
+All tests use synthetic data and run on CPU — no GPU or external files required.
+
+---
+
+## Code Conventions
+
+- **Imports** — all internal imports use `from customics.xxx import yyy` (never `from src.xxx`).
+- **Type hints** — add them to every public function and class signature.
+- **Docstrings** — NumPy style with `Parameters`, `Returns`, and `Raises` sections.
+  Add a short `Examples` block for public-facing methods.
+- **Logging** — use `logger = logging.getLogger(__name__)` instead of `print()`.
+- **No `eval()`** — use dict dispatch for dynamic object construction (see `_ACTIVATIONS` in `tasks/survival.py`).
+- **`nn.ModuleList`** — use it (not plain `list`) for collections of `nn.Module` objects.
+
+---
+
+## Submitting Changes
+
+1. Fork the repository and create a branch off `main`.
+2. Make your changes, add tests for new behaviour, and ensure the full test suite passes.
+3. Open a pull request against `main` with a clear description of what changed and why.
+
+---
+
+## Reporting Issues
+
+Please open an issue at [github.com/HakimBenkirane/CustOmics/issues](https://github.com/HakimBenkirane/CustOmics/issues) and include:
+- A minimal reproducible example
+- Your Python and PyTorch versions
+- The full traceback if applicable