kernelforge 0.1.2__tar.gz → 0.1.8__tar.gz

{kernelforge-0.1.2 → kernelforge-0.1.8}/.github/workflows/release.yaml

@@ -1,5 +1,5 @@
 # .github/workflows/release.yml
-name: Build & Publish
+name: "Build & Publish"
 
 on:
   release:
@@ -12,7 +12,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-22.04] # macos-14]
+        os: [ubuntu-22.04, macos-latest]
 
     steps:
       - uses: actions/checkout@v4
@@ -21,12 +21,24 @@ jobs:
         with:
           python-version: "3.12"   # host Python, cibuildwheel makes all others
 
+      - name: Clean wheelhouse and build artifacts
+        run: |
+          rm -rf dist/ build/ *.egg-info wheelhouse/
+
+      - name: Install gfortran (macOS)
+        if: runner.os == 'macOS'
+        run: brew install gcc libomp
+
+      - name: Check gfortran version
+        if: runner.os == 'macOS'
+        run: which gfortran-14
+
       - name: Install cibuildwheel
         run: python -m pip install cibuildwheel==2.*
 
       - name: Build wheels
         env:
-          CIBW_BUILD: "cp312-*"
+          CIBW_BUILD: "cp310-* cp311-* cp312-* cp313-*"
           CIBW_SKIP: "pp* *-musllinux_* cp*-manylinux_i686"
           CIBW_TEST_COMMAND: "pytest -q {project}/tests -k 'not slow' -x"
           CIBW_TEST_EXTRAS: "test"
@@ -47,6 +59,15 @@ jobs:
             CMAKE_ARGS
             OPENBLAS_NUM_THREADS
             OMP_NUM_THREADS
+          CIBW_ENVIRONMENT_MACOS: >
+            MACOSX_DEPLOYMENT_TARGET=15.0
+            CMAKE_ARGS="-DBLAS_VENDOR=Apple"
+            FC=gfortran-14
+          CIBW_ENVIRONMENT_PASS_MACOS: >
+            MACOSX_DEPLOYMENT_TARGET
+            CMAKE_ARGS
+            FC
+          CIBW_ARCHS_MACOS: arm64
         run: python -m cibuildwheel --output-dir wheelhouse
 
       - name: Build sdist
@@ -55,20 +76,32 @@ jobs:
       - name: Upload artifacts
         uses: actions/upload-artifact@v4
         with:
-          name: wheels
+          name: wheels-${{ runner.os }}
           path: wheelhouse/*
-
   publish:
     needs: build-wheels
     runs-on: ubuntu-22.04
-    if: github.event_name == 'release'
+    if: github.event_name == 'release' || github.event_name == 'workflow_dispatch'
     steps:
       - uses: actions/download-artifact@v4
         with:
-          name: wheels
+          name: wheels-Linux
           path: dist
+
+      - uses: actions/download-artifact@v4
+        with:
+          name: wheels-macOS
+          path: dist
+
+      - name: Flatten artifacts
+        run: |
+          mkdir -p dist/flat
+          find dist -name '*.whl' -exec cp {} dist/flat/ \;
+          find dist -name '*.tar.gz' -exec cp {} dist/flat/ \;
+
       - uses: pypa/gh-action-pypi-publish@v1.11.0
         with:
           user: __token__
           password: ${{ secrets.PYPI_API_TOKEN }}
+          packages-dir: dist/flat
 

kernelforge-0.1.8/Makefile

@@ -0,0 +1,8 @@
+install:
+	pip install -e .[test] --verbose
+
+test:
+	pytest
+
+environment:
+	conda env create -f environments/environment-dev.yaml

kernelforge-0.1.8/PKG-INFO

@@ -0,0 +1,114 @@
+Metadata-Version: 2.2
+Name: kernelforge
+Version: 0.1.8
+Summary: Optimzed Kernels for ML
+Author: Anders Christensen
+License: MIT
+Project-URL: Homepage, https://github.com/youruser/kernelforge
+Project-URL: Issues, https://github.com/youruser/kernelforge/issues
+Requires-Python: >=3.10
+Requires-Dist: numpy>=2.00
+Provides-Extra: test
+Requires-Dist: pytest>=8; extra == "test"
+Requires-Dist: pytest-xdist; extra == "test"
+Requires-Dist: pytest-cov; extra == "test"
+Requires-Dist: pytest-timeout; extra == "test"
+Description-Content-Type: text/markdown
+
+# KernelForge - Optimized Lernels for ML
+
+I really only care about writing optimized kernel code, so this project will be completed as I find additional time... XD 
+
+I'm reviving this project to finish an old project using random Fourier features for kernel ML.
+
+
+# Installation
+
+```bash
+conda env create -f environments/environment-dev.yml
+pip install -e .
+pytest -v -s
+```
+## PyPI installation
+
+Install the requirements (e.g. the conda env above) and install from PyPI.
+This should work on both MacOS and Linux/PC:
+
+```bash
+conda activate kernelforge-dev
+pip install kernelforge
+```
+This will install pre-compiled wheels with gfortran and linked againts OpenBLAS on Linux and Accelerate on MacOS.
+If you want to use MKL or other BLAS/LAPACK libraries, you need to compile from source, see below.
+
+
+## Intel compilers and MKL
+
+It is 2025 so you can `sudo apt get install intel-basekit` on Linux/PC to get the compilers and MKL.
+Then set up the environment variables:
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+In this case, MKL will be autodetected by some CMake magic. If you additionally want to compile with Intel compilers, you can set the environment variables when running `pip install`:
+```bash
+CC=icx CXX=icpx FC=ifx make install
+```
+
+## Timings
+
+| Function Name | QML [s] | Kernelforge [s] |
+|:---------------|------------:|--------------------:|
+| Upper triangle Gaussian kernel 16K x 16K| 1.82 | 0.64 |
+| Kernel Jacobian |  |  |
+| Kernel Hessian |  |  |
+
+## TODO list
+
+The goal is to remove pain-points of existing QML libraries
+- Improved use of BLAS/LAPACK routines
+- Removal of Fortran dependencies
+  - No Fortran-ordered arrays
+  - No Fortran compilers needed
+- Simplified build system
+  - No cooked F2PY/Meson build system
+- Simplified entrypoints that are compatible with RDKit, ASE, Scikit-learn, etc.
+  - A few high-level functions that do the most common tasks efficiently and correctly
+
+#### Todos:
+- Houskeeping:
+  - [x] Pybind11 bindings and CMake build system
+  - [x] Setup CI with GitHub Actions
+  - [ ] Rewrite existing kernels to C++ (no Fortran)
+  - [x] Setup GHA to build PyPI wheels
+  - [x] Test Linux build matrices
+  - [ ] Test MacOS build matrices
+  - [ ] Test Windows build matrices
+  - [ ] Add build for all Python version >=3.11
+- Ensure correct linking with optimized BLAS/LAPACK libraries:
+  - [x] OpenBLAS (Linux) <- also used in wheels
+  - [x] AMD BLIS and libflame (Linux)
+  - [x] MKL (Linux)
+  - [x] Accelerated (MacOS)
+- Add kernels:
+  - [x] Gaussian kernel
+  - [ ] Jacobian/gradient kernel
+  - [ ] Optimized Jacobian kernel for single inference
+  - [ ] Hessian kernel
+  - [ ] GDML-like kernel
+- Add random Fourier features kernel code
+  - [ ] RFF kernel
+  - [ ] RFF gradient kernel
+  - [ ] RFF chunked DSYRK kernel
+  - The same as above, just for Hadamard features when I find the time
+- Add standard solvers:
+  - [ ] Cholesky
+  - [ ] QR and/or SVD for non-square matrices
+- Add moleular descriptors with derivatives:
+  - [ ] Coulomb matrix
+  - [ ] FCHL19 + derivatives
+  - [ ] GDML-like inverse-distance matrix + derivatives
+- [ ] Plan structure for saving models for inference as `.npz` files
+#### Stretch goals:
+- [ ] Plan RDKit interface
+- [ ] Plan Scikit-learn interface
+- [ ] Plan ASE interface

kernelforge-0.1.8/README.md

@@ -0,0 +1,97 @@
+# KernelForge - Optimized Lernels for ML
+
+I really only care about writing optimized kernel code, so this project will be completed as I find additional time... XD 
+
+I'm reviving this project to finish an old project using random Fourier features for kernel ML.
+
+
+# Installation
+
+```bash
+conda env create -f environments/environment-dev.yml
+pip install -e .
+pytest -v -s
+```
+## PyPI installation
+
+Install the requirements (e.g. the conda env above) and install from PyPI.
+This should work on both MacOS and Linux/PC:
+
+```bash
+conda activate kernelforge-dev
+pip install kernelforge
+```
+This will install pre-compiled wheels with gfortran and linked againts OpenBLAS on Linux and Accelerate on MacOS.
+If you want to use MKL or other BLAS/LAPACK libraries, you need to compile from source, see below.
+
+
+## Intel compilers and MKL
+
+It is 2025 so you can `sudo apt get install intel-basekit` on Linux/PC to get the compilers and MKL.
+Then set up the environment variables:
+```bash
+source /opt/intel/oneapi/setvars.sh
+```
+In this case, MKL will be autodetected by some CMake magic. If you additionally want to compile with Intel compilers, you can set the environment variables when running `pip install`:
+```bash
+CC=icx CXX=icpx FC=ifx make install
+```
+
+## Timings
+
+| Function Name | QML [s] | Kernelforge [s] |
+|:---------------|------------:|--------------------:|
+| Upper triangle Gaussian kernel 16K x 16K| 1.82 | 0.64 |
+| Kernel Jacobian |  |  |
+| Kernel Hessian |  |  |
+
+## TODO list
+
+The goal is to remove pain-points of existing QML libraries
+- Improved use of BLAS/LAPACK routines
+- Removal of Fortran dependencies
+  - No Fortran-ordered arrays
+  - No Fortran compilers needed
+- Simplified build system
+  - No cooked F2PY/Meson build system
+- Simplified entrypoints that are compatible with RDKit, ASE, Scikit-learn, etc.
+  - A few high-level functions that do the most common tasks efficiently and correctly
+
+#### Todos:
+- Houskeeping:
+  - [x] Pybind11 bindings and CMake build system
+  - [x] Setup CI with GitHub Actions
+  - [ ] Rewrite existing kernels to C++ (no Fortran)
+  - [x] Setup GHA to build PyPI wheels
+  - [x] Test Linux build matrices
+  - [ ] Test MacOS build matrices
+  - [ ] Test Windows build matrices
+  - [ ] Add build for all Python version >=3.11
+- Ensure correct linking with optimized BLAS/LAPACK libraries:
+  - [x] OpenBLAS (Linux) <- also used in wheels
+  - [x] AMD BLIS and libflame (Linux)
+  - [x] MKL (Linux)
+  - [x] Accelerated (MacOS)
+- Add kernels:
+  - [x] Gaussian kernel
+  - [ ] Jacobian/gradient kernel
+  - [ ] Optimized Jacobian kernel for single inference
+  - [ ] Hessian kernel
+  - [ ] GDML-like kernel
+- Add random Fourier features kernel code
+  - [ ] RFF kernel
+  - [ ] RFF gradient kernel
+  - [ ] RFF chunked DSYRK kernel
+  - The same as above, just for Hadamard features when I find the time
+- Add standard solvers:
+  - [ ] Cholesky
+  - [ ] QR and/or SVD for non-square matrices
+- Add moleular descriptors with derivatives:
+  - [ ] Coulomb matrix
+  - [ ] FCHL19 + derivatives
+  - [ ] GDML-like inverse-distance matrix + derivatives
+- [ ] Plan structure for saving models for inference as `.npz` files
+#### Stretch goals:
+- [ ] Plan RDKit interface
+- [ ] Plan Scikit-learn interface
+- [ ] Plan ASE interface

kernelforge-0.1.8/environments/environment-dev.yaml

@@ -0,0 +1,29 @@
+name: kernelforge-dev
+channels:
+  - conda-forge
+  - defaults
+dependencies:
+  - python=3.12
+  - pip
+  - pandas
+  - mamba
+  - numpy
+  - plotly
+  - matplotlib
+  - jupyterlab
+  - ipykernel
+  - ipywidgets
+  - nbconvert
+  - tqdm
+  - black
+  - isort
+  - scipy
+  - scikit-learn
+  - seaborn
+  - flake8
+  - mypy
+  - pre-commit
+  - pytest
+  - pytest-cov
+  - pytest-xdist
+  - pytest-timeout

{kernelforge-0.1.2 → kernelforge-0.1.8}/pyproject.toml

@@ -4,14 +4,14 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "kernelforge"
-version = "0.1.2"
-description = "Fortran kernels with C ABI and Python bindings (CMake/skbuild/pybind11)"
+version = "0.1.8"
+description = "Optimzed Kernels for ML"
 readme = "README.md"
 license = {text = "MIT"}
 authors = [{name = "Anders Christensen"}]
-requires-python = ">=3.9"
+requires-python = ">=3.10"
 dependencies = [
-    "numpy>=2.00",   # required at runtime
+    "numpy>=2.00",
 ]
 
 [project.optional-dependencies]
@@ -23,14 +23,10 @@ Issues = "https://github.com/youruser/kernelforge/issues"
 
 [tool.scikit-build]
 wheel.expand-macos-universal-tags = true
-wheel.py-api = "py3"
 cmake.build-type = "Release"
 cmake.verbose = true
 wheel.packages = ["python/kernelforge"]
 
-# optional: put compiled outputs under build/{tag}/ to avoid clashes
-# build-dir = "build/{wheel_tag}"
-
 [tool.scikit-build.cmake.define]
 CMAKE_VERBOSE_MAKEFILE = "ON"
 CMAKE_BUILD_TYPE = "Release"

kernelforge-0.1.2/Makefile

@@ -1,5 +0,0 @@
-install:
-	pip install -e .[test] --verbose
-
-test:
-	pytest

kernelforge-0.1.2/PKG-INFO

@@ -1,41 +0,0 @@
-Metadata-Version: 2.2
-Name: kernelforge
-Version: 0.1.2
-Summary: Fortran kernels with C ABI and Python bindings (CMake/skbuild/pybind11)
-Author: Anders Christensen
-License: MIT
-Project-URL: Homepage, https://github.com/youruser/kernelforge
-Project-URL: Issues, https://github.com/youruser/kernelforge/issues
-Requires-Python: >=3.9
-Requires-Dist: numpy>=2.00
-Provides-Extra: test
-Requires-Dist: pytest>=8; extra == "test"
-Requires-Dist: pytest-xdist; extra == "test"
-Requires-Dist: pytest-cov; extra == "test"
-Requires-Dist: pytest-timeout; extra == "test"
-Description-Content-Type: text/markdown
-
-# kernelforge
-Optimized kernels for ML
-
-- Without using F2PY or Meson
-
-# Installation
-
-```bash
-pip install -e .
-pytest -v -s
-```
-
-## Intel compilers and MKL
-
-GNU compilers will be used by default. If you want to use Intel compilers and MKL, you can set the environment variables:
-
-```bash
-source /opt/intel/oneapi/setvars.sh
-```
-In this case, MKL will be autodetected and used. If you additionally want to compile with Intel compilers, you can set the environment variables when running `pip install`:
-
-```bash
-CC=icx CXX=icpx FC=ifx make install
-```

kernelforge-0.1.2/README.md

@@ -1,24 +0,0 @@
-# kernelforge
-Optimized kernels for ML
-
-- Without using F2PY or Meson
-
-# Installation
-
-```bash
-pip install -e .
-pytest -v -s
-```
-
-## Intel compilers and MKL
-
-GNU compilers will be used by default. If you want to use Intel compilers and MKL, you can set the environment variables:
-
-```bash
-source /opt/intel/oneapi/setvars.sh
-```
-In this case, MKL will be autodetected and used. If you additionally want to compile with Intel compilers, you can set the environment variables when running `pip install`:
-
-```bash
-CC=icx CXX=icpx FC=ifx make install
-```