code review

Author: gpmp-dev

README.md

@@ -1,21 +1,16 @@
-# GPmp: the Gaussian process micro package
+# GPmp: the Gaussian Process micro package
 
-## What is GPmp?
+GPmp is a lightweight toolkit for Gaussian process (GP) modeling. It
+provides the essential components for GP-based algorithms with a focus
+on speed and customization.
 
-GPmp provides simple building blocks for GP-based algorithms.  It is meant to
-be fast and easily customizable.
+## Features
 
-The user can choose between three backends to perform numerical
-computations: plain numpy, [JAX](https://jax.readthedocs.io/) or
-[PyTorch](). When JAX or PyTorch is used, GPmp relies on
-auto-differentiation to compute gradients. GPmp also uses JIT
-compilation features provided by JAX.
-
-As core features, GPmp implements:
- * GP interpolation and regression with known or unknown mean / intrinsinc kriging
- * The standard Gaussian likelihood and the restricted likelihood of a model
- * Leave-one-out predictions using fast cross-validation formulas
- * Conditional sample paths
+- **GP interpolation & regression:** Supports known or unknown mean
+    functions (including intrinsic kriging).
+- **Likelihood computation:** standard Gaussian and restricted likelihood.
+- **Efficient cross-validation:** fast leave-one-out predictions.
+- **Conditional sampling:** generate conditional sample paths.
 
 It is up to the user to write the mean and covariance functions for
 setting a GP model.
@@ -28,101 +23,88 @@ However, for the purpose of the example, GPmp provides functions for:
    likelihood, or user-defined criteria
  * model diagnosis
  * vizualization of results
+ * ...
+
+## Backends
 
-Note also that for the purpose of simplicity, the gpmp does not check
-the validity of arguments. This is left to the responsibility of the
-user / calling code.
+GPmp supports three numerical backends:
+- **PyTorch:** Dynamic computation with auto-differentiation.
+- **JAX:** Auto-differentiation with JIT compilation.
+- **NumPy:** Basic computation (default if neither PyTorch nor JAX are
+    found).
+
+On startup, GPmp automatically selects the backend in this order:
+PyTorch → JAX → NumPy. You can override this by setting the
+`GPMP_BACKEND` environment variable before launching GPmp.
 
 ## Installation
 
-Clone the git repository:
-```
+Clone the repository:
+```bash
 git clone https://github.com/gpmp-dev/gpmp.git
 ```
-(or download a zip version)
-
-Install in dev mode:
-```
+Install in development mode:
+```bash
 pip install -e .
 ```
 
-Remarks:
-
-1. It is recommended to run GPmp using PyTorch or JAX as a backend instead of Numpy.
-
-2. Upon initialization, GPmp automatically detects if PyTorch is
-   installed and sets the GPMP_BACKEND environment variable to 'torch'
-   by default. If PyTorch is not found, it then searches for JAX and
-   sets GPMP_BACKEND to 'jax'. If neither PyTorch nor JAX are
-   detected, GPmp defaults to using Numpy, setting the GPMP_BACKEND
-   environment variable to 'numpy'. Users have the option to manually
-   override the chosen backend by setting the GPMP_BACKEND variable
-   before launching GPmp.
-   
-2. JAX, which requires jaxlib, which is
-   supported on Linux and macOS platforms, and on Windows via the
-   Windows Subsystem for Linux. There is some initial native Windows
-   support, but it is still somewhat immature (see below).
-
-2. To install JAX with both CPU and NVidia GPU support, CUDA and CuDNN
-   must be installed first. See
-   https://github.com/google/jax#installation and
-   https://jax.readthedocs.io/en/latest/changelog.html for details.
-
-   For instance, to install jaxlib  with Cuda 11 and cudnn 8.2 or newer:
-```bash   
-pip install jax[cuda11_cudnn82] -f https://storage.googleapis.com/jax-releases/jax_releases.html
-```
-   
-3. To install jaxlib on Windows, one can follow instructions at
-   https://github.com/cloudhan/jax-windows-builder.  Select and
-   download a version of jaxlib from
-   https://whls.blob.core.windows.net/unstable/index.html according to
-   your Python version. Then install jax manually.
-
-```powershell
-pip install <jaxlib_whl>
-pip install jax
-```
+### Backend Setup
+
+- **PyTorch:** Recommended for best performance.
+- **JAX:** 
+  - For CPU-only usage, install directly from PyPI:
+    ```bash
+    pip install jax
+    ```
+  - For NVIDIA GPU support (e.g., with CUDA 12), run:
+    ```bash
+    pip install -U "jax[cuda12]"
+    ```
+  Detailed, platform-specific instructions are available in the
+  [official JAX documentation](https://github.com/google/jax#installation).
 
 ## Usage
 
-Please refer to the examples that illustrate the different features of GPmp.
+Check the examples in the repository for a quick start. Customize your
+own mean and covariance functions as needed.
 
-## Documentation with Sphinx
-The automatic documentation is created using [Sphinx](https://www.sphinx-doc.org/en/master/) and following the python docstrings recommendations from [numpydoc](https://numpydoc.readthedocs.io/en/latest/format.html)
+## Documentation
 
-The html theme is taken from
->pip install pydata_sphinx_theme
+GPmp’s documentation is built with
+[Sphinx](https://www.sphinx-doc.org/en/master/) using the PyData
+theme. To generate the HTML documentation:
 
-In order to generate the intereactive html:
->cd ./doc
-
->make html
+```bash
+cd doc
+make html
+```
 
-## To do
+## How to Cite
 
-* write documentation
+If you use GPmp in your research, please cite it as follows:
 
-## Authors
+```bibtex
+@software{gpmp2025,
+  author       = {Emmanuel Vazquez},
+  title        = {GPmp: the Gaussian Process micro package},
+  year         = {2025},
+  url          = {https://github.com/gpmp-dev/gpmp},
+  note         = {Version x.y},
+}
+```
 
- See AUTHORS.md
+*Please update the version number as appropriate.*
 
-## Copyright
+## Future Work
 
- Copyright (C) 2022-2024 CentraleSupelec
+- Expand documentation and tutorials.
+- Enhance diagnostic tools and model visualization.
 
-## License
+## Authors
 
- GPmp is free software: you can redistribute it and/or modify it
- under the terms of the GNU General Public License as published by
- the Free Software Foundation, either version 3 of the License, or
- (at your option) any later version.
+See [AUTHORS.md](AUTHORS.md) for details.
 
- GPmp is distributed in the hope that it will be useful, but WITHOUT
- ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
- or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
- License for more details.
+## License
 
- You should have received a copy of the GNU General Public License
- along with gpmp. If not, see http://www.gnu.org/licenses/.
+GPmp is free software released under the GNU General Public License v3.0.
+See [LICENSE](LICENSE.txt) for more details.

VERSION

@@ -1 +1 @@
-0.9.18
+0.9.19

changelog.md

@@ -1,3 +1,4 @@
+- 0.9.19 introduce 'DifferentiableFunction' to encapsulate gradient computation, code enhancement
 - 0.9.18 small improvements in kernel.py and num.py
 - 0.9.17 refactor parameter selection procedures for more general selection criteria support
 - 0.9.16 changes in core.py, num.py about  Cholesky factorization

examples/gpmp_example02_1d_interpolation.py

@@ -2,7 +2,7 @@
 Plot and optimize the restricted negative log-likelihood
 
 Author: Emmanuel Vazquez <[email protected]>
-Copyright (c) 2022-2024, CentraleSupelec
+Copyright (c) 2022-2025, CentraleSupelec
 License: GPLv3 (see LICENSE)
 """
 

examples/gpmp_example03_2d.py

@@ -14,7 +14,7 @@
 
 ----
 Author: Emmanuel Vazquez <[email protected]>
-Copyright (c) 2022-2024, CentraleSupelec
+Copyright (c) 2022-2025, CentraleSupelec
 License: GPLv3 (see LICENSE)
 ----
 This example is based on the file stk_example_kb03.m from the STK at

examples/gpmp_example04_nd.py

@@ -13,7 +13,7 @@
 
 ----
 Author: Emmanuel Vazquez <[email protected]>
-Copyright (c) 2022-2024, CentraleSupelec
+Copyright (c) 2022-2025, CentraleSupelec
 License: GPLv3 (see LICENSE)
 """
 import gpmp.num as gnp

examples/gpmp_example06_1d_regression.py

@@ -4,7 +4,7 @@
 
 ----
 Author: Emmanuel Vazquez <[email protected]>
-Copyright (c) 2022-2024, CentraleSupelec
+Copyright (c) 2022-2025, CentraleSupelec
 License: GPLv3 (see LICENSE)
 ----
 This example is based on the file stk_example_kb01.m from the STK at

examples/gpmp_example20_1d_interpolation_variation_remap.py

@@ -2,7 +2,7 @@
 Plot and optimize the restricted negative log-likelihood
 
 Author: Emmanuel Vazquez <[email protected]>
-Copyright (c) 2022-2024, CentraleSupelec
+Copyright (c) 2022-2025, CentraleSupelec
 License: GPLv3 (see LICENSE)
 """
 

examples/gpmp_example21_1d_interpolation_variation_reml.py

@@ -2,7 +2,7 @@
 Plot and optimize the restricted negative log-likelihood
 
 Author: Emmanuel Vazquez <[email protected]>
-Copyright (c) 2022-2024, CentraleSupelec
+Copyright (c) 2022-2025, CentraleSupelec
 License: GPLv3 (see LICENSE)
 """
 
@@ -82,10 +82,12 @@ def main():
     covparam0 = gp.kernel.anisotropic_parameters_initial_guess(model, xi, zi)
 
     # selection criterion
+    selection_criterion = gp.kernel.negative_log_restricted_likelihood
     nlrl, dnlrl = gp.kernel.make_selection_criterion_with_gradient(
-        model, gp.kernel.negative_log_restricted_likelihood, xi, zi
+        model, selection_criterion, xi, zi
     )
 
+    # optimize parameters
     covparam_reml, info = gp.kernel.autoselect_parameters(
         covparam0, nlrl, dnlrl, silent=False, info=True
     )

examples/gpmp_example22_1d_interpolation_variation_ml.py

@@ -2,7 +2,7 @@
 Plot and optimize the restricted negative log-likelihood
 
 Author: Emmanuel Vazquez <[email protected]>
-Copyright (c) 2022-2024, CentraleSupelec
+Copyright (c) 2022-2025, CentraleSupelec
 License: GPLv3 (see LICENSE)
 """