scikit-ntk 1.1.0.dev0

Implementation of the neural tangent kernel for scikit-learn's Gaussian process module.

Neural Tangent Kernel for scikit-learn Gaussian Processes

scikit-ntk is implementation of the neural tangent kernel (NTK) for the scikit-learn machine learning library as part of "An Empirical Analysis of the Laplace and Neural Tangent Kernels" master's thesis (found at http://hdl.handle.net/20.500.12680/d504rr81v and https://arxiv.org/abs/2208.03761). This library is meant to directly integrate with sklearn.gaussian_process module. This implementation of the NTK can be used in combination with other kernels to train and predict with Gaussian process regressors and classifiers.

Installation

Dependencies

scikit-ntk requires:

• Python (>=3.7)
• scikit-learn (>=1.0.1)

User installation

In terminal using pip run:

pip install scikit-ntk

Usage

Usage is described in examples/usage.py; however, to get started simply import the NeuralTangentKernel class:

from skntk import NeuralTangentKernel as NTK

kernel_ntk = NTK(D=3, bias=0.01, bias_bounds=(1e-6, 1e6))

Once declared, usage is the same as other scikit-learn kernels.

Building

Python Poetry (>=1.2) is required if you wish to build scikit-ntk from source. In order to build follow these steps:

1. Clone the repository

git clone [e-mail]:392781/scikit-ntk.git

2. Enable a Poetry virtual environment

poetry shell

3. Build and install

poetry build
poetry install --with dev

Citation

If you use scikit-ntk in your scientific work, please use the following citation alongside the scikit-learn citations found at https://scikit-learn.org/stable/about.html#citing-scikit-learn:

@mastersthesis{lencevicius2022laplacentk,
  author  = "Ronaldas Paulius Lencevicius",
  title   = "An Empirical Analysis of the Laplace and Neural Tangent Kernels",
  school  = "California State Polytechnic University, Pomona",
  year    = "2022",
  month   = "August",
  note    = {\url{http://hdl.handle.net/20.500.12680/d504rr81v}}
}