Update scvi 1x

.github/workflows/build.yaml

@@ -15,10 +15,10 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v3
-      - name: Set up Python 3.10
+      - name: Set up Python 3.12
         uses: actions/setup-python@v4
         with:
-          python-version: "3.10"
+          python-version: "3.12"
           cache: "pip"
           cache-dependency-path: "**/pyproject.toml"
       - name: Install build dependencies

.github/workflows/test.yaml

@@ -23,10 +23,10 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - os: ubuntu-latest
-            python: "3.10"
           - os: ubuntu-latest
             python: "3.12"
+          - os: ubuntu-latest
+            python: "3.13"
           - os: ubuntu-latest
             python: "3.12"
             pip-flags: "--pre"

.readthedocs.yaml

@@ -1,9 +1,9 @@
 # https://docs.readthedocs.io/en/stable/config-file/v2.html
 version: 2
 build:
-  os: ubuntu-20.04
+  os: ubuntu-22.04
   tools:
-    python: "3.10"
+    python: "3.12"
 sphinx:
   configuration: docs/conf.py
   # disable this for more lenient docs builds

CHANGELOG.md

@@ -8,46 +8,61 @@ and this project adheres to [Semantic Versioning][].
 [keep a changelog]: https://keepachangelog.com/en/1.0.0/
 [semantic versioning]: https://semver.org/spec/v2.0.0.html
 
+## Unreleased
+
+### Changed
+
+- Migrated to scvi-tools ≥1.4 (Lightning 2.x): replaced `use_gpu` with `accelerator`/`devices`, updated `SaveBestState` → `SaveCheckpoint`, `_initialize_model`, and `_get_loaded_data` call sites.
+- `GroupDataSplitter` now splits at the **sample level** (not cell level) using `external_indexing`, ensuring every validation bag is complete.
+- Bumped `requires-python` to `>=3.12`; switched package manager to [uv](https://docs.astral.sh/uv/) with a committed `uv.lock`.
+- numpy ≥2.0 and anndata ≥0.12 compatibility fixes (dtype identity check, pandas `groupby(observed=True)`).
+
+### Fixed
+
+- Single-layer regressor head used `z_dim` instead of `class_input_dim` when sample covariates are present.
+- `.squeeze()` on 1-element tensors in `create_df` produced 0-d scalars; wrapped with `np.atleast_1d`.
+- Fixed floating-point undercount in `GroupDataSplitter` (`floor(1.0 - 0.9) = 0` edge case).
+
 ## [0.3.2] - 2025-12-16
 
 ### Added
 
--   tutorial for regression
--   support for sample-covariate embeddings, experimental, only one-hot
+- tutorial for regression
+- support for sample-covariate embeddings, experimental, only one-hot
 
 ### Changed
 
--   removed `muon` from dependencies for tutorials
--   changed the classification tutorial to include training across 3 CV folds
+- removed `muon` from dependencies for tutorials
+- changed the classification tutorial to include training across 3 CV folds
 
-## Fixed
+### Fixed
 
--   fixed a bug in regression which was caused by a wrong key when accessing the continuous covariates registered with scvi-tools
+- fixed a bug in regression which was caused by a wrong key when accessing the continuous covariates registered with scvi-tools
 
 ## [0.3.1] - 2025-07-14
 
 ### Fixed
 
--   Fixed a bug in `score_top_cells` that didn't set the specified `key_added` column in .obs to True.
--   Fixed a bug in `score_top_cells` that set the `key_added` to be categorical, which resulted in wrong indexing in `get_sample_representations`.
+- Fixed a bug in `score_top_cells` that didn't set the specified `key_added` column in .obs to True.
+- Fixed a bug in `score_top_cells` that set the `key_added` to be categorical, which resulted in wrong indexing in `get_sample_representations`.
 
 ## [0.3.0] - 2025-07-13
 
 ### Added
 
--   **Utility functions**: Added `score_top_cells` and `get_sample_representations` to utils module
--   `score_top_cells`: Function to identify and score top cells based on attention weights
--   `get_sample_representations`: Function to aggregate cell-level data to sample-level representations
+- **Utility functions**: Added `score_top_cells` and `get_sample_representations` to utils module
+- `score_top_cells`: Function to identify and score top cells based on attention weights
+- `get_sample_representations`: Function to aggregate cell-level data to sample-level representations
 
 ### Changed
 
--   **Major refactoring**: Removed MultiVAE and MultiVAE_MIL models, keeping only MIL classifier
--   **Code cleanup**: Removed MLP attention weight learning from Aggregator class
--   **Parameter consistency**: Fixed default values between model and module classes
--   **Dynamic z_dim**: Automatically infer z_dim from input data shape instead of hardcoded value
+- **Major refactoring**: Removed MultiVAE and MultiVAE_MIL models, keeping only MIL classifier
+- **Code cleanup**: Removed MLP attention weight learning from Aggregator class
+- **Parameter consistency**: Fixed default values between model and module classes
+- **Dynamic z_dim**: Automatically infer z_dim from input data shape instead of hardcoded value
 
 ### Fixed
 
--   **Make categorical covariates categorical**: Ensure the correct type in `setup_anndata`.
--   **Improved error handling**: If the prediction covariate hasn't been registered with `setup_anndata`, throw an error.
--   **Dead links to API and changelog**: Fixed in README.
+- **Make categorical covariates categorical**: Ensure the correct type in `setup_anndata`.
+- **Improved error handling**: If the prediction covariate hasn't been registered with `setup_anndata`, throw an error.
+- **Dead links to API and changelog**: Fixed in README.

README.md

@@ -12,38 +12,35 @@
 
 Please refer to the [documentation][link-docs]. In particular, the
 
--   [API documentation][link-api]
+- [API documentation][link-api]
 
 and the tutorials:
 
--   [Classification with MultiMIL](https://multimil.readthedocs.io/en/latest/notebooks/mil_classification.html) [![Open In Colab][badge-colab]](https://colab.research.google.com/github/theislab/multimil/blob/main/docs/notebooks/mil_classification.ipynb)
--   [Regression with MultiMIL](https://multimil.readthedocs.io/en/latest/notebooks/mil_regression.html) [![Open In Colab][badge-colab]](https://colab.research.google.com/github/theislab/multimil/blob/main/docs/notebooks/mil_regression.ipynb)
+- [Classification with MultiMIL](https://multimil.readthedocs.io/en/latest/notebooks/mil_classification.html) [![Open In Colab][badge-colab]](https://colab.research.google.com/github/theislab/multimil/blob/main/docs/notebooks/mil_classification.ipynb)
+- [Regression with MultiMIL](https://multimil.readthedocs.io/en/latest/notebooks/mil_regression.html) [![Open In Colab][badge-colab]](https://colab.research.google.com/github/theislab/multimil/blob/main/docs/notebooks/mil_regression.ipynb)
 
 Please also check out our [sample prediction pipeline](https://github.com/theislab/sample-prediction-pipeline), which contains MultiMIL and several other baselines.
 
 ## Installation
 
-You need to have Python 3.10 or newer installed on your system. We recommend installing [Mambaforge](https://github.com/conda-forge/miniforge#mambaforge).
+You need to have Python 3.12 or newer installed on your system. We recommend using [uv](https://docs.astral.sh/uv/) for environment management.
 
-To create and activate a new environment:
+Install the latest release of `multimil` from [PyPI][link-pypi]:
 
 ```bash
-mamba create --name multimil python=3.10
-mamba activate multimil
+uv pip install multimil
 ```
 
-Next, there are several alternative options to install multimil:
-
-1. Install the latest release of `multimil` from [PyPI][link-pypi]:
+Or install the latest development version:
 
 ```bash
-pip install multimil
+uv pip install git+https://github.com/theislab/multimil.git@main
 ```
 
-2. Or install the latest development version:
+Alternatively, with plain pip:
 
 ```bash
-pip install git+https://github.com/theislab/multimil.git@main
+pip install multimil
 ```
 
 ## Release notes

docs/api.md

@@ -23,3 +23,31 @@
 
     module.MILClassifierTorch
 ```
+
+## Neural network components
+
+```{eval-rst}
+.. module:: multimil.nn
+.. currentmodule:: multimil
+
+.. autosummary::
+    :toctree: generated
+
+    nn.MLP
+    nn.Aggregator
+```
+
+## Utilities
+
+```{eval-rst}
+.. module:: multimil.utils
+.. currentmodule:: multimil
+
+.. autosummary::
+    :toctree: generated
+
+    utils.get_sample_representations
+    utils.score_top_cells
+    utils.setup_ordinal_regression
+    utils.plt_plot_losses
+```

docs/contributing.md

@@ -10,10 +10,18 @@ to the [scanpy developer guide][].
 ## Installing dev dependencies
 
 In addition to the packages needed to _use_ this package, you need additional python packages to _run tests_ and _build
-the documentation_. It's easy to install them using `pip`:
+the documentation_. We use [uv](https://docs.astral.sh/uv/) for dependency management:
 
 ```bash
 cd multimil
+uv sync --extra dev --extra test --extra doc
+```
+
+This creates a `.venv/` and installs all dependencies from `uv.lock`. Prefix commands with `uv run` (e.g. `uv run pytest`) to use the managed environment.
+
+Alternatively, with plain pip:
+
+```bash
 pip install -e ".[dev,test,doc]"
 ```
 
@@ -51,7 +59,7 @@ and [prettier][prettier-editors].
 ## Writing tests
 
 ```{note}
-Remember to first install the package with `pip install -e '.[dev,test]'`
+Remember to first install the package with `uv sync --extra dev --extra test` (or `pip install -e '.[dev,test]'`).
 ```
 
 This package uses the [pytest][] for automated testing. Please [write tests][scanpy-test-docs] for every function added
@@ -61,7 +69,7 @@ Most IDEs integrate with pytest and provide a GUI to run tests. Alternatively, y
 command line by executing
 
 ```bash
-pytest
+uv run pytest
 ```
 
 in the root of the repository.
@@ -99,11 +107,11 @@ Specify `vX.X.X` as a tag name and create a release. For more information, see [
 
 Please write documentation for new or changed features and use-cases. This project uses [sphinx][] with the following features:
 
--   the [myst][] extension allows to write documentation in markdown/Markedly Structured Text
--   [Numpy-style docstrings][numpydoc] (through the [napoloen][numpydoc-napoleon] extension).
--   Jupyter notebooks as tutorials through [myst-nb][] (See [Tutorials with myst-nb](#tutorials-with-myst-nb-and-jupyter-notebooks))
--   [Sphinx autodoc typehints][], to automatically reference annotated input and output types
--   Citations (like {cite:p}`Virshup_2023`) can be included with [sphinxcontrib-bibtex](https://sphinxcontrib-bibtex.readthedocs.io/)
+- the [myst][] extension allows to write documentation in markdown/Markedly Structured Text
+- [Numpy-style docstrings][numpydoc] (through the [napoloen][numpydoc-napoleon] extension).
+- Jupyter notebooks as tutorials through [myst-nb][] (See [Tutorials with myst-nb](#tutorials-with-myst-nb-and-jupyter-notebooks))
+- [Sphinx autodoc typehints][], to automatically reference annotated input and output types
+- Citations (like {cite:p}`Virshup_2023`) can be included with [sphinxcontrib-bibtex](https://sphinxcontrib-bibtex.readthedocs.io/)
 
 See the [scanpy developer docs](https://scanpy.readthedocs.io/en/latest/dev/documentation.html) for more information
 on how to write documentation.
@@ -120,10 +128,10 @@ repository.
 
 #### Hints
 
--   If you refer to objects from other packages, please add an entry to `intersphinx_mapping` in `docs/conf.py`. Only
-    if you do so can sphinx automatically create a link to the external documentation.
--   If building the documentation fails because of a missing link that is outside your control, you can add an entry to
-    the `nitpick_ignore` list in `docs/conf.py`
+- If you refer to objects from other packages, please add an entry to `intersphinx_mapping` in `docs/conf.py`. Only
+  if you do so can sphinx automatically create a link to the external documentation.
+- If building the documentation fails because of a missing link that is outside your control, you can add an entry to
+  the `nitpick_ignore` list in `docs/conf.py`
 
 #### Building the docs locally