Getting started

Installation

Install from source (current development version):

pip install -e .

Basic usage

The central concept in svdynamics is the CompositeKernel object.

A composite kernel represents a weighted sum of multiple base kernels.

Example:

from svdynamics import CompositeKernel, SVDClassifier
from sklearn.datasets import make_classification

X, y = make_classification(
    n_samples=300,
    n_features=10,
    random_state=0
)

kernel = CompositeKernel(
    kernels=[
        ("rbf", {"gamma": 0.2}),
        ("linear", {}),
        ("poly", {"degree": 2, "coef0": 1.0}),
    ],
    weights=[0.6, 0.3, 0.1],
    normalize=True,
)

clf = SVDClassifier(
    C=1.0,
    kernel=kernel,
    probability=True,
    random_state=0,
)

clf.fit(X, y)

y_pred = clf.predict(X)
y_prob = clf.predict_proba(X)

How it works

Composite kernels are implemented as callable kernels passed directly to scikit-learn’s SVC / SVR.

The resulting kernel matrix is computed as a weighted sum of the individual kernel matrices:

\[K(x, x') = \sum_{i=1}^{m} w_i \, K_i(x, x')\]

This makes svdynamics fully compatible with sklearn internals.