From 379932ce397c88f87b4f193a272f1c87a1da5b77 Mon Sep 17 00:00:00 2001 From: jessegrabowski Date: Fri, 29 May 2026 17:02:05 -0500 Subject: [PATCH 1/4] Build sdist/wheel with build-and-inspect action for pymc-devs org --- .github/workflows/pypi.yml | 18 ++++-------------- 1 file changed, 4 insertions(+), 14 deletions(-) diff --git a/.github/workflows/pypi.yml b/.github/workflows/pypi.yml index 257cd69..5f072a6 100644 --- a/.github/workflows/pypi.yml +++ b/.github/workflows/pypi.yml @@ -9,29 +9,19 @@ on: jobs: build: - name: build source distribution + name: build and inspect the package runs-on: ubuntu-latest steps: - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1 with: fetch-depth: 0 - - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0 - with: - python-version: "3.12" - - name: Build the sdist and the wheel - run: | - pip install build - python3 -m build - - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2 - with: - name: artifact - path: dist/* + - uses: hynek/build-and-inspect-python-package@d44ca7d91762de7a7d5436ddae667c6da6d1c3df # v2.18.0 publish: name: upload release to PyPI needs: [build] runs-on: ubuntu-latest - if: github.repository_owner == 'bwengals' && github.event_name == 'release' && github.event.action == 'published' + if: github.repository_owner == 'pymc-devs' && github.event_name == 'release' && github.event.action == 'published' # Use the `release` GitHub environment to protect the Trusted Publishing (OIDC) # workflow by requiring signoff from a maintainer. environment: release @@ -41,6 +31,6 @@ jobs: steps: - uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0 with: - name: artifact + name: Packages path: dist - uses: pypa/gh-action-pypi-publish@cef221092ed1bacb1cc03d23a2d87d1d172e277b # v1.14.0 From 4a82113ddb05437c55d7084e4aa442082df69ca8 Mon Sep 17 00:00:00 2001 From: jessegrabowski Date: Fri, 29 May 2026 17:04:21 -0500 Subject: [PATCH 2/4] Update repository URLs, authorship, and copyright for pymc-devs org --- LICENSE | 2 +- README.md | 8 ++++---- pyproject.toml | 7 ++++--- 3 files changed, 9 insertions(+), 8 deletions(-) diff --git a/LICENSE b/LICENSE index 94ab251..cf4aa8a 100644 --- a/LICENSE +++ b/LICENSE @@ -1,6 +1,6 @@ MIT License -Copyright (c) 2025 Bill Engels +Copyright (c) 2025 Bill Engels and the PyMC Developers Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal diff --git a/README.md b/README.md index 16638e9..008f4db 100644 --- a/README.md +++ b/README.md @@ -12,7 +12,7 @@ PTGP is for practitioners who need flexible, well-supported GP modeling. The goa - **PyMC priors:** set priors on any hyperparameter; use PyMC distributions for mean functions and noise models; MAP training by default - **Training tools:** L-BFGS-B and Adam optimizers, per-parameter learning rates, staged optimization, frozen variables, inducing point initialization strategies, diagnostic-guided workflows; more are being added, such as carefully monitored training to help diagnose issues early - **Agent-readable docs:** `docs/agents/` ships LLM-readable guides for debugging training issues and folk wisdom (VFE training covered). See the [Working with AI coding assistants](#working-with-ai-coding-assistants) section below. -- **More coming:** see the [issues](https://github.com/bwengals/ptgp/issues) +- **More coming:** see the [issues](https://github.com/pymc-devs/ptgp/issues) Researchers benefit from the underlying design: PTGP is built on PyTensor's symbolic graph and rewrite system, so you write GP math directly (`pt.linalg.inv(K)`, `pt.linalg.slogdet(K)`) and the compiler chooses efficient algorithms based on declared matrix structure. This makes it straightforward to implement new GP approximations and create custom models, and will eventually allow matrix structure like Kronecker, Toeplitz, and sparse to be taken advantage of automatically. @@ -95,17 +95,17 @@ python scripts/install_claude_skills.py --user # ~/.claude/skills/ ## Install ```bash -pip install git+https://github.com/bwengals/ptgp.git +pip install git+https://github.com/pymc-devs/ptgp.git ``` To hack on PTGP itself, clone and install in editable mode: ```bash -git clone https://github.com/bwengals/ptgp.git +git clone https://github.com/pymc-devs/ptgp.git cd ptgp pip install -e . ``` ## Contributing -See the [issues](https://github.com/bwengals/ptgp/issues) for what's being worked on. Feel free to propose issues, feature requests, or use cases you've been hoping could be made easier. PRs always welcome. +See the [issues](https://github.com/pymc-devs/ptgp/issues) for what's being worked on. Feel free to propose issues, feature requests, or use cases you've been hoping could be made easier. PRs always welcome. diff --git a/pyproject.toml b/pyproject.toml index 7a1e176..ff5d54c 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -5,7 +5,8 @@ build-backend = "hatchling.build" [project] name = "ptgp" authors = [ - {name = 'Bill Engels', email = 'w.j.engels@gmail.com'} + {name = 'Bill Engels', email = 'w.j.engels@gmail.com'}, + {name = 'PyMC Developers', email = 'pymc.devs@gmail.com'} ] description = "A practitioner's toolbox for estimating large-scale Gaussian Process models with PyMC and PyTensor" requires-python = ">=3.12" @@ -31,8 +32,8 @@ dependencies = [ dev = ["pytest", "pre-commit", "mypy", "polars"] [project.urls] -Repository = "https://github.com/bwengals/ptgp.git" -Issues = "https://github.com/bwengals/ptgp/issues" +Repository = "https://github.com/pymc-devs/ptgp.git" +Issues = "https://github.com/pymc-devs/ptgp/issues" [tool.hatch.version] source = 'vcs' From 8f1900760caac3ba8a6bad0bcede503712ce007b Mon Sep 17 00:00:00 2001 From: jessegrabowski Date: Fri, 29 May 2026 17:05:03 -0500 Subject: [PATCH 3/4] Fix broken lower-level example block and Claude Code header in README --- README.md | 19 +++++++++++++++---- 1 file changed, 15 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index 008f4db..2140769 100644 --- a/README.md +++ b/README.md @@ -54,13 +54,22 @@ with pm.Model() as model: mean, var = pg.predict(svgp, np.linspace(-3, 3, 100)[:, None], fit) ``` -`pg.fit` picks a default objective from the gp type (`Unapproximated` → `marginal_log_likelihood`, `VFE` → `collapsed_elbo`, `SVGP` → `elbo`) and returns a `FitResult` that `pg.predict` consumes. For stochastic mini-batch training, staged VFE, or per-group learning rates, drop down to `pg.optim.compile_training_step` / `pg.optim.compile_scipy_objective` — see [`notebooks/demo.ipynb`](notebooks/demo.ipynb). +`pg.fit` picks a default objective from the gp type (`Unapproximated` → `marginal_log_likelihood`, `VFE` → `collapsed_elbo`, `SVGP` → `elbo`) and returns a `FitResult` that `pg.predict` consumes. For stochastic mini-batch training, staged VFE, or per-group learning rates, drop down to `pg.optim.compile_training_step` / `pg.optim.compile_scipy_objective` — see [`notebooks/demo.ipynb`](notebooks/demo.ipynb): + +```python +X_var = pt.matrix("X") +y_var = pt.vector("y") + +step, shared_params, shared_extras = pg.optim.compile_training_step( + pg.objectives.elbo, svgp, X_var, y_var, model, learning_rate=1e-2 +) for i in range(500): loss = step(X, y) -predict_fn = pg.optim.compile_predict(svgp, pt.matrix("X_new"), model, shared_params, - extra_vars=vp.extra_vars, shared_extras=shared_extras) +predict_fn = pg.optim.compile_predict( + svgp, pt.matrix("X_new"), model, shared_params, shared_extras=shared_extras +) mean, var = predict_fn(np.linspace(-3, 3, 100)[:, None]) ``` @@ -79,7 +88,9 @@ PTGP is set up to work nicely with AI coding assistants: - **[`AGENTS.md`](AGENTS.md)** — project-level instructions for AI coding assistants (architecture, conventions, where things live, how to run tests). Follows the [AGENTS.md](https://agents.md/) cross-tool convention used by Codex, Cursor, Aider, and others. - **[`docs/agents/`](docs/agents/)** — backend-agnostic agent-skill docs covering folk wisdom and training-debug recipes. Currently includes [`ptgp-vfe`](docs/agents/ptgp-vfe/) (VFE diagnostic skill: pitfalls, escalation workflow, interpretation of `VFEDiagnostics` and `GreedyVarianceDiagnostics`). -**Claude Code users:** Claude Code reads `CLAUDE.md`, not `AGENTS.md`. Symlink so they stay in sync: +### Claude Code users + +Claude Code reads `CLAUDE.md`, not `AGENTS.md`. Symlink so they stay in sync: ```bash ln -s AGENTS.md CLAUDE.md From f132f665440a3623d7356dbf3214b282633ee055 Mon Sep 17 00:00:00 2001 From: jessegrabowski Date: Fri, 29 May 2026 17:11:14 -0500 Subject: [PATCH 4/4] Add readme field so twine check --strict passes --- pyproject.toml | 1 + 1 file changed, 1 insertion(+) diff --git a/pyproject.toml b/pyproject.toml index ff5d54c..0351f64 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -9,6 +9,7 @@ authors = [ {name = 'PyMC Developers', email = 'pymc.devs@gmail.com'} ] description = "A practitioner's toolbox for estimating large-scale Gaussian Process models with PyMC and PyTensor" +readme = "README.md" requires-python = ">=3.12" keywords = [ "gaussian processes",