Created
July 15, 2018 15:32
-
-
Save lan496/3ce0fe7ece96f06039b4462b32c59360 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import numpy as np | |
from scipy.special import iv | |
from sklearn.base import BaseEstimator, TransformerMixin | |
from sklearn.utils.validation import check_is_fitted | |
class ExpSineSquaredSampler(BaseEstimator, TransformerMixin): | |
""" | |
Approximates feature map of an Exp-Sine-Squared kernel by its Fourier series expansion. | |
Exp-Sine-Squared kernel is given by: | |
k(x, y) = exp(-2(sin(pi * d(x, y) / periodicity) / length_scale)^2) | |
Parameters | |
---------- | |
length_scale : float | |
periodicity : float | |
degrees : int | |
the number of incorporated terms | |
Attributes | |
---------- | |
weights_ : array, (degrees, ) | |
coef : array, (degrees, ) | |
""" | |
def __init__(self, length_scale=1., periodicity=1., degrees=30): | |
self.length_scale = length_scale | |
self.periodicity = periodicity | |
self.degrees = degrees | |
def fit(self, X=None, y=None): | |
""" | |
Parameters | |
---------- | |
Returns | |
------- | |
self | |
""" | |
self.weights_ = np.arange(1, self.degrees + 1) * 2 * np.pi / self.periodicity | |
self.coef_ = iv(np.arange(1, self.degrees + 1), self.length_scale ** (-2)) \ | |
/ np.exp(self.length_scale ** (-2)) | |
self.coef_ = np.sqrt(self.coef_) | |
return self | |
def transform(self, X): | |
""" | |
Parameters | |
---------- | |
X : (n_samples, [n_features]) | |
Returns | |
------- | |
X_new : (n_samples, 2 * degrees * n_features) | |
""" | |
check_is_fitted(self, 'weights_') | |
if X.ndim == 1: | |
X = X[:, np.newaxis] | |
n_samples, n_features = X.shape | |
projection = X[:, :, np.newaxis] * self.weights_[np.newaxis, np.newaxis, :] | |
X_new = np.concatenate([self.coef_[np.newaxis, np.newaxis, :] * np.cos(projection), | |
self.coef_[np.newaxis, np.newaxis, :] * np.sin(projection)], | |
axis=2) | |
X_new = np.reshape(X_new, (n_samples, n_features * 2 * self.degrees)) | |
return X_new |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment