athena.feature_map.rff_jac

rff_jac(inputs, pr_matrix, bias, n_features, sigma_f)

Random Fourier Features map’s Jacobian. It can be vectorized for inputs of shape n_samples-by-input_dim.

Parameters

• inputs (numpy.ndarray) – input_dim dimensional inputs to project to the RKHS.

• pr_matrix (numpy.ndarray) – n_features-by-input_dim projection matrix, whose rows are sampled from the spectral distribution.

• bias (numpy.ndarray) – n_features dimensional bias. It is sampled from a Unifrom distribution on the interval [0, 2*PI].

• n_features (int) – dimension of the RKHS

• sigma_f (int) – multiplicative term representing the empirical variance the outptus.

Returns

n_features-by-input_dim dimensional projection of the inputs to the RKHS

Return type

numpy.ndarray