AlphaPFN 0.0.1__tar.gz

alphapfn-0.0.1/.gitignore

@@ -0,0 +1,29 @@
+__pycache__/
+*.pyc
+*.pyo
+*.pyd
+.Python
+*.egg-info/
+*.egg
+dist/
+build/
+
+# checkpoints (workspace local; eventually fetched from HF Hub)
+*.pt
+*.pth
+*.safetensors
+final_models/
+checkpoints/
+
+# uv venv
+.venv/
+
+# wandb / experiment tracking
+wandb/
+
+# editors
+.vscode/
+.idea/
+
+# notebooks
+.ipynb_checkpoints/

alphapfn-0.0.1/CITATION.cff

@@ -0,0 +1,69 @@
+cff-version: 1.2.0
+message: "If you use this software, please cite the paper below."
+title: "alphapfn: Fast Entropy Search via In-Context Learning"
+abstract: >-
+  Lightweight package for α-PFN, a Prior-Fitted Network that amortizes
+  information-theoretic acquisition functions (PES, MES, JES) for
+  Bayesian Optimization. α-PFN replaces the slow Monte-Carlo sampling
+  used by classical entropy-search methods with a single transformer
+  forward pass.
+type: software
+version: 0.0.1
+license: MIT
+repository-code: "https://github.com/automl/AlphaPFN"
+url: "https://github.com/automl/AlphaPFN"
+keywords:
+  - bayesian-optimization
+  - entropy-search
+  - prior-fitted-networks
+  - pfn
+  - in-context-learning
+  - amortized-inference
+  - acquisition-function
+authors:
+  - family-names: "Rakotoarison"
+    given-names: "Herilalaina"
+    email: "rakoheri@ad.helsinki.fi"
+    affiliation: "University of Helsinki"
+  - family-names: "Adriaensen"
+    given-names: "Steven"
+    email: "adriaens@cs.uni-freiburg.de"
+    affiliation: "University of Freiburg"
+  - family-names: "Viering"
+    given-names: "Tom"
+    email: "t.j.viering@tudelft.nl"
+    affiliation: "Delft University of Technology"
+  - family-names: "Hvarfner"
+    given-names: "Carl"
+    affiliation: "Meta"
+  - family-names: "Müller"
+    given-names: "Samuel"
+    affiliation: "Meta"
+  - family-names: "Hutter"
+    given-names: "Frank"
+    affiliation: "Prior Labs / ELLIS Institute Tübingen / University of Freiburg"
+  - family-names: "Bakshy"
+    given-names: "Eytan"
+    affiliation: "Meta"
+preferred-citation:
+  type: conference-paper
+  title: "α-PFN: Fast Entropy Search via In-Context Learning"
+  year: 2026
+  url: "https://openreview.net/forum?id=7Oonij8oLU"
+  conference:
+    name: "Forty-third International Conference on Machine Learning (ICML)"
+  authors:
+    - family-names: "Rakotoarison"
+      given-names: "Herilalaina"
+    - family-names: "Adriaensen"
+      given-names: "Steven"
+    - family-names: "Viering"
+      given-names: "Tom"
+    - family-names: "Hvarfner"
+      given-names: "Carl"
+    - family-names: "Müller"
+      given-names: "Samuel"
+    - family-names: "Hutter"
+      given-names: "Frank"
+    - family-names: "Bakshy"
+      given-names: "Eytan"

alphapfn-0.0.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 AutoML-Freiburg-Hannover
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

alphapfn-0.0.1/PKG-INFO

@@ -0,0 +1,160 @@
+Metadata-Version: 2.4
+Name: AlphaPFN
+Version: 0.0.1
+Summary: Fast Entropy Search via In-Context Learning (PFN-based amortized acquisition functions for Bayesian Optimization)
+Project-URL: Homepage, https://github.com/automl/AlphaPFN
+Project-URL: Repository, https://github.com/automl/AlphaPFN
+Project-URL: Issues, https://github.com/automl/AlphaPFN/issues
+Project-URL: Paper, https://openreview.net/forum?id=7Oonij8oLU
+Author: Herilalaina Rakotoarison, Steven Adriaensen, Tom Viering, Carl Hvarfner, Samuel Müller, Frank Hutter, Eytan Bakshy
+Maintainer-email: Herilalaina Rakotoarison <rkt.herilalaina@gmail.com>
+License: MIT License
+        
+        Copyright (c) 2026 AutoML-Freiburg-Hannover
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+License-File: LICENSE
+Keywords: acquisition-function,amortized-inference,bayesian-optimization,entropy-search,in-context-learning,pfn,prior-fitted-networks,pytorch
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Scientific/Engineering :: Mathematics
+Requires-Python: >=3.10
+Requires-Dist: einops
+Requires-Dist: gpytorch>=1.5.0
+Requires-Dist: networkx
+Requires-Dist: numpy<2.0,>=1.21.2
+Requires-Dist: packaging
+Requires-Dist: platformdirs
+Requires-Dist: safetensors
+Requires-Dist: torch>=1.9.0
+Requires-Dist: typing-extensions
+Provides-Extra: botorch
+Requires-Dist: botorch>=0.15.0; extra == 'botorch'
+Description-Content-Type: text/markdown
+
+# $\alpha$-PFN: Fast Entropy Search via In-Context Learning
+
+[![PyPI version](https://img.shields.io/pypi/v/AlphaPFN.svg)](https://pypi.org/project/AlphaPFN/)
+[![Python](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+[![Paper](https://img.shields.io/badge/Paper-OpenReview-b31b1b.svg)](https://openreview.net/forum?id=7Oonij8oLU)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/automl/AlphaPFN/blob/main/examples/quickstart.ipynb)
+
+**$\alpha$-PFN** is a Prior-Fitted Network that amortizes information-theoretic acquisition functions. Supported acquisition functions: Predictive Entropy Search (PES), Max-value Entropy Search (MES), and Joint Entropy Search (JES). 
+
+<p align="center">
+  <img src="images/hero.gif" alt="Traditional GP-based Entropy Search samples optima via RFF and averages conditional entropies over N MC samples; α-PFN approximates the same acquisition in a single transformer forward pass.">
+</p>
+
+> To reproduce our ICML 2026 paper experiments, see branch
+> [`icml2026`](https://github.com/automl/AlphaPFN/tree/icml2026).
+
+## Install
+
+```bash
+pip install AlphaPFN
+```
+
+Or from source:
+
+```bash
+git clone https://github.com/automl/AlphaPFN
+cd AlphaPFN
+uv sync
+```
+
+Pretrained checkpoints (~20 MB) download automatically on the first `from_pretrained` call and cache under `~/.cache/alphapfn/`.
+
+## Quick start
+
+A self-contained 6D BO loop on Hartmann, using `botorch.optim.optimize_acqf`. Requires the `[botorch]` extra — `pip install "AlphaPFN[botorch]"`:
+
+```python
+import torch
+from botorch.optim import optimize_acqf
+from botorch.test_functions import Hartmann
+from alphapfn import AlphaPFN
+
+# 1. Objective on the unit cube (α-PFN maximizes — `negate=True` flips Hartmann's sign).
+hartmann = Hartmann(dim=6, negate=True)
+
+# 2. Initial design.
+torch.manual_seed(0)
+d, n_init, num_steps = 6, 6, 30
+X = torch.rand(n_init, d, dtype=torch.double)
+y = hartmann(X)
+bounds = torch.stack([torch.zeros(d), torch.ones(d)]).double()
+
+# 3. Load the pretrained acquisition; checkpoints download on first call.
+acqf = AlphaPFN.from_pretrained(acquisition="JES")
+
+# 4. BO loop.
+for step in range(num_steps):
+    acqf.fit(X, y)                            # fit() standardizes y internally
+    X_next, _ = optimize_acqf(acqf, bounds=bounds, q=1,
+                              num_restarts=10, raw_samples=128)
+    y_next = hartmann(X_next.squeeze(0))
+    X = torch.cat([X, X_next.detach().double()])
+    y = torch.cat([y, y_next.detach().double().reshape(1)])
+    print(f"step {step+1:>2}: best so far = {y.max().item():.4f}")
+```
+
+Runnable version: [`examples/bo_with_optimize_acqf.py`](examples/bo_with_optimize_acqf.py)
+or open the [Colab notebook](https://colab.research.google.com/github/automl/AlphaPFN/blob/main/examples/quickstart.ipynb).
+
+## API
+
+```python
+AlphaPFN.from_pretrained(
+    acquisition: str | None = None,   # "PES" (default), "MES", or "JES"
+    version: str = "v1",
+    *,
+    load_base_model: bool = False,
+    ucb_beta: float = 2.0,
+    strict: bool = True,              # pass strict=False to skip input checks
+)
+```
+
+Before fitting, prepare your data so that:
+- **You are maximizing.** To minimize instead, negate your objective.
+  This is NOT checked, so forgetting it silently gives wrong results.
+- **Each input feature lies in `[0, 1]`.** Rescale your search space accordingly.
+
+`fit()` standardizes targets internally (`standardize_y=True` by default) — pass raw $y$. Pass `standardize_y=False` if you have already standardized. `strict=True` (default) validates the unit-cube contract on every `fit`/`forward`; pass `strict=False` to skip.
+
+## Cite
+
+```bibtex
+@inproceedings{
+  rakotoarison2026alphapfn,
+  title={{$\alpha$}-PFN: Fast Entropy Search via In-Context Learning},
+  author={Rakotoarison, Herilalaina and Adriaensen, Steven and Viering, Tom and Hvarfner, Carl and M{\"u}ller, Samuel and Hutter, Frank and Bakshy, Eytan},
+  booktitle={Forty-third International Conference on Machine Learning},
+  year={2026},
+  url={https://openreview.net/forum?id=7Oonij8oLU}
+}
+```

alphapfn-0.0.1/README.md

@@ -0,0 +1,102 @@
+# $\alpha$-PFN: Fast Entropy Search via In-Context Learning
+
+[![PyPI version](https://img.shields.io/pypi/v/AlphaPFN.svg)](https://pypi.org/project/AlphaPFN/)
+[![Python](https://img.shields.io/badge/python-3.10%2B-blue.svg)](https://www.python.org/downloads/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
+[![Paper](https://img.shields.io/badge/Paper-OpenReview-b31b1b.svg)](https://openreview.net/forum?id=7Oonij8oLU)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/automl/AlphaPFN/blob/main/examples/quickstart.ipynb)
+
+**$\alpha$-PFN** is a Prior-Fitted Network that amortizes information-theoretic acquisition functions. Supported acquisition functions: Predictive Entropy Search (PES), Max-value Entropy Search (MES), and Joint Entropy Search (JES). 
+
+<p align="center">
+  <img src="images/hero.gif" alt="Traditional GP-based Entropy Search samples optima via RFF and averages conditional entropies over N MC samples; α-PFN approximates the same acquisition in a single transformer forward pass.">
+</p>
+
+> To reproduce our ICML 2026 paper experiments, see branch
+> [`icml2026`](https://github.com/automl/AlphaPFN/tree/icml2026).
+
+## Install
+
+```bash
+pip install AlphaPFN
+```
+
+Or from source:
+
+```bash
+git clone https://github.com/automl/AlphaPFN
+cd AlphaPFN
+uv sync
+```
+
+Pretrained checkpoints (~20 MB) download automatically on the first `from_pretrained` call and cache under `~/.cache/alphapfn/`.
+
+## Quick start
+
+A self-contained 6D BO loop on Hartmann, using `botorch.optim.optimize_acqf`. Requires the `[botorch]` extra — `pip install "AlphaPFN[botorch]"`:
+
+```python
+import torch
+from botorch.optim import optimize_acqf
+from botorch.test_functions import Hartmann
+from alphapfn import AlphaPFN
+
+# 1. Objective on the unit cube (α-PFN maximizes — `negate=True` flips Hartmann's sign).
+hartmann = Hartmann(dim=6, negate=True)
+
+# 2. Initial design.
+torch.manual_seed(0)
+d, n_init, num_steps = 6, 6, 30
+X = torch.rand(n_init, d, dtype=torch.double)
+y = hartmann(X)
+bounds = torch.stack([torch.zeros(d), torch.ones(d)]).double()
+
+# 3. Load the pretrained acquisition; checkpoints download on first call.
+acqf = AlphaPFN.from_pretrained(acquisition="JES")
+
+# 4. BO loop.
+for step in range(num_steps):
+    acqf.fit(X, y)                            # fit() standardizes y internally
+    X_next, _ = optimize_acqf(acqf, bounds=bounds, q=1,
+                              num_restarts=10, raw_samples=128)
+    y_next = hartmann(X_next.squeeze(0))
+    X = torch.cat([X, X_next.detach().double()])
+    y = torch.cat([y, y_next.detach().double().reshape(1)])
+    print(f"step {step+1:>2}: best so far = {y.max().item():.4f}")
+```
+
+Runnable version: [`examples/bo_with_optimize_acqf.py`](examples/bo_with_optimize_acqf.py)
+or open the [Colab notebook](https://colab.research.google.com/github/automl/AlphaPFN/blob/main/examples/quickstart.ipynb).
+
+## API
+
+```python
+AlphaPFN.from_pretrained(
+    acquisition: str | None = None,   # "PES" (default), "MES", or "JES"
+    version: str = "v1",
+    *,
+    load_base_model: bool = False,
+    ucb_beta: float = 2.0,
+    strict: bool = True,              # pass strict=False to skip input checks
+)
+```
+
+Before fitting, prepare your data so that:
+- **You are maximizing.** To minimize instead, negate your objective.
+  This is NOT checked, so forgetting it silently gives wrong results.
+- **Each input feature lies in `[0, 1]`.** Rescale your search space accordingly.
+
+`fit()` standardizes targets internally (`standardize_y=True` by default) — pass raw $y$. Pass `standardize_y=False` if you have already standardized. `strict=True` (default) validates the unit-cube contract on every `fit`/`forward`; pass `strict=False` to skip.
+
+## Cite
+
+```bibtex
+@inproceedings{
+  rakotoarison2026alphapfn,
+  title={{$\alpha$}-PFN: Fast Entropy Search via In-Context Learning},
+  author={Rakotoarison, Herilalaina and Adriaensen, Steven and Viering, Tom and Hvarfner, Carl and M{\"u}ller, Samuel and Hutter, Frank and Bakshy, Eytan},
+  booktitle={Forty-third International Conference on Machine Learning},
+  year={2026},
+  url={https://openreview.net/forum?id=7Oonij8oLU}
+}
+```

alphapfn-0.0.1/RELEASING.md

@@ -0,0 +1,67 @@
+# Releasing AlphaPFN to PyPI
+
+Steps for a new release (maintainers only).
+
+## Prereqs (one-time)
+
+1. Have a PyPI account with maintainer rights to the `AlphaPFN` package
+   (or TestPyPI for dry runs).
+2. Create an API token in your PyPI account settings.
+3. Install build tooling:
+
+```bash
+pip install --upgrade build twine
+```
+
+## Cut a release
+
+1. Bump the version in two places:
+   - `pyproject.toml` → `version = "X.Y.Z"`
+   - `alphapfn/__init__.py` → `__version__ = "X.Y.Z"`
+
+2. Tag and push:
+
+```bash
+git commit -am "Release X.Y.Z"
+git tag vX.Y.Z
+git push && git push --tags
+```
+
+3. Build the distribution artifacts (cleans `dist/` first):
+
+```bash
+rm -rf dist/ build/ *.egg-info
+python -m build
+```
+
+This produces `dist/alphapfn-X.Y.Z-py3-none-any.whl` and the sdist.
+
+4. (Optional) Test on TestPyPI first:
+
+```bash
+twine upload --repository testpypi dist/*
+pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ AlphaPFN==X.Y.Z
+```
+
+5. Upload to PyPI:
+
+```bash
+twine upload dist/*
+```
+
+6. Verify the install works in a clean env:
+
+```bash
+python -m venv /tmp/alpha-pfn-check && source /tmp/alpha-pfn-check/bin/activate
+pip install "AlphaPFN[botorch]==X.Y.Z"
+python -c "from alphapfn import AlphaPFN, __version__; print(__version__)"
+```
+
+7. Create a GitHub release pointing at the tag, with a short changelog.
+
+## Notes
+
+- Checkpoints are downloaded at runtime by `alphapfn.checkpoints`, so they
+  don't need to be bundled in the wheel. `pyproject.toml` already excludes
+  `*.pt`, `*.gz`, `images/`, and `examples/*.ipynb` from the build.
+- Test the wheel locally before publishing: `pip install dist/alphapfn-*.whl`.

alphapfn-0.0.1/alphapfn/__init__.py

@@ -0,0 +1,17 @@
+"""alphapfn — fast entropy-search acquisition via in-context learning.
+
+Public surface:
+    from alphapfn import AlphaPFN
+    model = AlphaPFN.from_pretrained(acquisition="JES")
+    model.fit(train_X, train_Y)
+    acq = model(X_test)
+"""
+from alphapfn.api import AlphaPFN, AlphaPFNPosteriorMean, ALLOWED_ACQUISITIONS
+
+__version__ = "0.0.1"
+__all__ = [
+    "AlphaPFN",
+    "AlphaPFNPosteriorMean",
+    "ALLOWED_ACQUISITIONS",
+    "__version__",
+]