release(0.2.0): scmidas.integrate() + bundled quickstart + Colab notebook

.gitignore

@@ -18,5 +18,8 @@ auto_ver.sh
 *.yaml
 *.png
 *.jpg
+# whitelist images committed to the docs static dir
+!docs/source/_static/img/*.png
+!docs/source/_static/img/*.jpg
 .venv*
 .nfs*

README.md

@@ -13,6 +13,7 @@
   <a href="https://pypi.org/project/scmidas/"><img src="https://img.shields.io/pypi/v/scmidas" alt="PyPI version"></a>
   <a href="https://scmidas.readthedocs.io/en/latest/"><img src="https://img.shields.io/readthedocs/scmidas" alt="Documentation Status"></a>
   <a href="https://github.com/labomics/midas/blob/main/LICENSE"><img src="https://img.shields.io/github/license/labomics/midas?v=1" alt="License"></a>
+  <a href="https://colab.research.google.com/github/labomics/midas/blob/main/examples/quickstart.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"></a>
   <a href="https://github.com/labomics/midas/stargazers"><img src="https://img.shields.io/github/stars/labomics/midas?style=social" alt="GitHub Stars"></a>
 </p>
 
@@ -38,26 +39,23 @@ pip install scmidas
 
 ## Quick start
 
-The MIDAS workflow is four calls. The snippet below is an API sketch — replace `...` with your data and refer to the [tutorials](https://scmidas.readthedocs.io/en/latest/) for runnable end-to-end examples.
+A bundled 1600-cell PBMC RNA+ADT mosaic dataset lets you go from `pip install` to a UMAP in about a minute on a single GPU — no extra downloads, no config files. Click the [Colab badge](https://colab.research.google.com/github/labomics/midas/blob/main/examples/quickstart.ipynb) to run it without installing anything, or copy the snippet:
 
 ```python
-from scmidas.config import load_config
-from scmidas.model import MIDAS
+import scmidas
 
-# 1. Build a model bound to a mosaic dataset.
-#    Input is either a directory of per-batch MTX matrices, or a MuData
-#    object via MIDAS.configure_data_from_mdata(...).
-model = MIDAS.configure_data_from_dir(load_config(), ..., transform={'atac': 'binarize'})
+mdata = scmidas.datasets.quickstart()      # bundled toy MuData
+model = scmidas.integrate(mdata)           # ~1 min on a mid-range GPU
+out   = model.predict(joint_latent=True)   # latent embeddings per batch
+```
 
-# 2. Train.
-model.train(max_epochs=2000)
+This produces lineage-separated clusters that mix freely across batches:
 
-# 3. Predict — latent embeddings (z_c, z_u) and imputed counts per batch.
-pred = model.predict()
+<div align="center">
+  <img src="docs/source/_static/img/quickstart_umap.png" alt="quickstart UMAP" width="850px">
+</div>
 
-# 4. (Optional) UMAP of the integrated latent space.
-model.get_emb_umap()
-```
+> ⚠️ **`scmidas.integrate()` defaults are tuned for the bundled toy dataset.** For your own data, override `max_epochs` (1000-2000 is typical) and consider letting `batch_size` default back to 256, e.g. `scmidas.integrate(my_mdata, max_epochs=2000, batch_size=256)`. See the [full demos](https://scmidas.readthedocs.io/en/latest/) for end-to-end pipelines on real-sized data, including imputation, batch correction, and cross-modality translation.
 
 ## Reproducibility
 

docs/source/release.md

@@ -4,6 +4,44 @@ All notable changes to this project will be documented in this file.
 
 ---
 
+## Version 0.2.x
+
+### v0.2.0 (2026-05-03)
+*   **🚀 New — `scmidas.integrate(mdata)` one-line entry point**
+    *   A thin top-level wrapper around `MIDAS.configure_data_from_mdata`
+        + `train()` with toy-tuned defaults (`batch_size=128`,
+        `max_epochs=65`, `lr=3e-4`) so that the bundled quickstart
+        dataset converges in roughly one minute on a single mid-range
+        GPU. The full `MIDAS` class API is unchanged for users who
+        need control.
+    *   ⚠️ The defaults are tuned for the toy quickstart only. For
+        real datasets, override `max_epochs` (1000-2000) and consider
+        `batch_size=256`.
+*   **🚀 New — bundled quickstart dataset**
+    *   `scmidas.datasets.quickstart()` returns a 1600-cell PBMC RNA+ADT
+        mosaic MuData (4 batches, full mosaic structure: one RNA-only,
+        one ADT-only, two paired). 500 RNA HVGs + 224 ADT features,
+        2.66 MB shipped inside the wheel.
+    *   Source: hand-tuned subset of `wnn_mosaic_8batch_mtx`. Build
+        script: `scripts/build_quickstart_demo.py`.
+*   **📚 Documentation**
+    *   New `examples/quickstart.ipynb` — pre-rendered notebook that
+        users can open in Colab via the new badge in the README, no
+        local install required.
+    *   README quickstart rewritten: replaces the previous `...` API
+        sketch with a runnable five-line snippet using
+        `scmidas.datasets.quickstart()` + `scmidas.integrate()`,
+        followed by the rendered UMAP image.
+*   **⚙️ Packaging**
+    *   `pyproject.toml` ships `data/*.h5mu` as package data so the
+        quickstart dataset travels with the wheel.
+    *   Module-level `logging.basicConfig(level=INFO)` removed from
+        five files (`config`, `data`, `model`, `nn`, `utils`); each
+        now does the canonical `logger = logging.getLogger(__name__)`
+        instead. Demo notebooks call `logging.basicConfig` themselves
+        so visible output is unchanged. Libraries should not call
+        `basicConfig` — it overrides the user's own logging config.
+
 ## Version 0.1.x
 
 ### v0.1.19 (2026-05-03)