Getting Started

Once Falkon is installed, getting started is easy. The basic setup to use the Falkon estimator only requires few lines of code:

import torch
from sklearn.datasets import load_boston
from falkon import Falkon, kernels

X, Y = load_boston(return_X_y=True)
X = torch.from_numpy(X)
Y = torch.from_numpy(Y).reshape(-1, 1)

kernel = kernels.GaussianKernel(sigma=1.0)
model = Falkon(
    kernel=kernel,
    penalty=1e-6,
    M=100,
)
model.fit(X, Y)
predictions = model.predict(X)

Passing Options

A number of different options exist for both the Falkon and LogisticFalkon estimators (see falkon.FalkonOptions). All options can be passed to the estimator through the FalkonOptions class, like so:

from falkon import FalkonOptions, Falkon, kernels

# Options to: increase the amount of output information; avoid using the KeOps library
options = FalkonOptions(debug=True, keops_active="no")
kernel = kernels.GaussianKernel(sigma=1.0)

model = Falkon(kernel=kernel,
               penalty=1e-6,
               M=100,
               maxiter=10,   # Set the maximum number of conjugate gradient iterations to 10
               options=options)

More Examples

For more detailed examples, have a look at the example notebooks.