torchkm 4.2.1

A PyTorch-based library for kernel machines (SVM, DWD, kernel logistic, etc.)

TorchKM: Fast Kernel Machines in PyTorch

TorchKM is a PyTorch-based library for kernel machines with a focus on fast train + tune workflows.

It currently provides:

• Kernel classification: kernel SVM, kernel DWD, and kernel logistic regression
• Fast model selection: pathwise solutions over a grid of regularization values (λ)
• Exact LOOCV support for kernel SVM
• GPU acceleration via PyTorch/CUDA, with safe CPU fallback
• A scikit-learn–style API for easy integration into existing Python workflows

Why TorchKM?

Kernel methods are still a strong choice when you want nonlinear decision boundaries, convex training objectives, and competitive performance on tabular or moderate-scale datasets. In practice, the bottleneck is often not training one model — it is training and tuning many models.

TorchKM is built for that workflow.

Installation

Minimal install

pip install torchkm

Install the scikit-learn wrapper

pip install "torchkm[sklearn]"

Development install

git clone https://github.com/YikaiZhang95/torchkm.git
cd torchkm
pip install -e ".[dev,sklearn]"
pytest -q

Quickstart

sklearn-style wrapper (recommended)

import numpy as np
import torch
from sklearn.datasets import make_circles
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

from torchkm.estimators import TorchKMSVC

# Toy nonlinear classification task
X, y = make_circles(n_samples=1200, factor=0.4, noise=0.08, random_state=0)
X = StandardScaler().fit_transform(X)
y = np.where(y == 0, -1, 1)   # TorchKM accepts {-1, +1} labels

Xtr, Xte, ytr, yte = train_test_split(X, y, test_size=0.2, random_state=0)

nfolds = 5
Cs = np.logspace(3, -3, num=12)

clf = TorchKMSVC(
    kernel="rbf",
    nC=len(Cs),
    Cs=Cs,
    nfolds=nfolds,
    device='cuda',
    probability=True,
    max_iter=200,
)

clf.fit(Xtr, ytr)

print("best lambda:", clf.best_C_)
print("test accuracy:", (clf.predict(Xte) == yte).mean())
print("first 3 probabilities:\n", clf.predict_proba(Xte[:3]))

Low-rank approximation

Use the low_rank=True when you want to handle a large data set.

clf = TorchKMSVC(
    kernel="rbf",
    low_rank=True,
    num_landmarks=500,
    nys_k=250,
    nC=20,
    nfolds=5,
    device='cuda',
    probability=True,
    platt_device='cuda',
)

clf.fit(Xtr, ytr)
(clf.predict(Xte) == yte).mean()

What TorchKM provides

High-level wrappers

• TorchKMSVC — kernel SVM classifier
• TorchKMDWD — kernel DWD classifier
• TorchKMLogit — kernel logistic regression

Common methods:

• fit(X, y)
• decision_function(X)
• predict(X)
• predict_proba(X) (when probability=True)
• platt_plot(X, y)

Low-level solvers

• cvksvm — kernel SVM with cross-validation over λ
• cvkdwd — kernel DWD with cross-validation over λ
• cvklogit — kernel logistic regression with cross-validation over λ

Utilities

• rbf_kernel, kernelMult, sigest

Device behavior

For portability, prefer:

device = "cuda" if torch.cuda.is_available() else "cpu"

TorchKM is designed to run on GPU when CUDA is available and fall back safely to CPU otherwise.

When to use TorchKM

TorchKM is a good fit when you want:

• nonlinear kernel classifiers without leaving the PyTorch ecosystem
• fast model selection across many regularization values
• exact LOOCV support for kernel SVM
• a wrapper API that feels familiar if you already use scikit-learn
• lower-level access to kernel matrices and solver internals

Testing

pytest -q

Contributing

Issues, bug reports, benchmarks, documentation improvements, and pull requests are welcome.

Good first contributions include:

• additional kernels
• multiclass wrappers
• more benchmark scripts
• expanded examples and tutorials

Citation

If you use TorchKM in academic work, please cite:

@article{zhang2026torchkm,
  title   = {TorchKM: GPU-Accelerated Kernel Machines with Fast Model Selection in PyTorch},
  author  = {Zhang, Yikai and Jia, Gaoxiang and Wang, Boxiang},
  journal = {Journal of Machine Learning Research (MLOSS track)},
  year    = {2026},
  note    = {Software paper submission}
}

License

MIT License. See LICENSE.

Contact

Yikai Zhang
[e-mail]