rfflearn.cpu.RFFGPC

Gaussian Process Classification with random Fourier features. This class is essentially the same as RFFGPR, but some pre-processing and post-processing are necessary for modifying a regression task to a classification task. The required processing is:

Assumed input label is vectors of class indexes, but the input of the RFFGPR should be one hot vectors of the class indexes.
Output of the RFFGPR is log probability, not predicted class indexes.

The purpose of this RFFGPC class is only to do these pre/post-processing.

def rfflearn.cpu.RFFGPC(self, dim_kernel=128, std_kernel=0.1, std_error=0.1, W=None, b=None, a=None, S=None)

Parameters:

dim_kernel: int  (default=128)

Dimension of the random matrix.

std_kernel: float  (default=0.1)

Standard deviation of the random matrix.

std_error: float  (default=0.1)

Standard deviation of the measurement error.

W: np.ndarray  (default=None)

Random matrix for the input X. If None then generated automatically.

b: np.ndarray  (default=None)

Random bias for the input X. If None then generated automatically.

a: np.ndarray  (default=None)

Cache of the matrix a. If None then generated automatically.

S: np.ndarray  (default=None)

Cache of the matrix S. If None then generated automatically.

Attributes:

dim: int

Dimension of the random matrix.

s_k: float

Standard deviation of the random matrix.

mat: Callable

A function to generate the random matrix W.

W: np.ndarray

Random matrix for the input X.

b: np.ndarray

Random bias for the input X.

s_e: np.ndarray

Standard deviation of the measurement error.

a: np.ndarray

Cache of the matrix a. If None then generated automatically.

S: np.ndarray

Cache of the matrix S. If None then generated automatically.

Member functions

def fit(self, X, y)

Trains the GPC model according to the given data.

Parameters:

X: np.ndarray

Input matrix with shape (n_samples, n_features_input).

y: np.ndarray

Output vector with shape (n_samples,).

args: dict

Extra arguments. However, this arguments will be ignored. This argument exists only for keeping the same interface with scikit-learn.

Returns:

rfflearn.cpu.RFFGPC

Fitted estimator.

def predict(self, X, return_std=False, return_cov=False)

Performs classification on the given data.

Parameters:

X: np.ndarray

Input matrix with shape (n_samples, n_features_input).

return_std: bool  (default=False)

Returns standard deviation of the prediction if True.

return_cov: bool  (default=False)

Returns covariance of the prediction if True.

Returns:

np.ndarray, or tuple

Predicted class indices vector, or tuple of prediction, standard deviation, and covariance of the prediction.

def score(self, X, y, **args)

Returns the mean accuracy on the given data and labels.

Parameters:

X: np.ndarray

Input matrix with shape (n_samples, n_features_input).

y: np.ndarray

True class label vector with shape (n_samples,).

args: dict  (default={})

Extra arguments. However, this arguments will be ignored.

Returns:

float

Mean classification accuracy.

Minimal Example

>>> import numpy as np                                  # Import Numpy
>>> import rfflearn.cpu as rfflearn                     # Import module
>>> X = np.array([[-1, -1], [-2, -1], [1, 1], [2, 1]])  # Define input data
>>> y = np.array([1, 1, 2, 2])                          # Defile label data
>>> gpc = rfflearn.RFFGPC().fit(X, y)                   # Training (on CPU)
>>> gpc.score(X, y)                                     # Inference (on CPU)
1.0
>>> gpc.predict(np.array([[-0.8, -1]]))
array([1])