dimensionality_simulations.py 文件源码

def mk_data(n_samples=200, random_state=0, separability=1,
            noise_corr=2, dim=100):
    rng = np.random.RandomState(random_state)
    y = rng.random_integers(0, 1, size=n_samples)
    noise = rng.normal(size=(n_samples, dim))
    if not noise_corr is None and noise_corr > 0:
        noise = ndimage.gaussian_filter1d(noise, noise_corr, axis=0)
    noise = noise / noise.std(axis=0)
    # We need to decrease univariate separability as dimension increases
    centers = 4. / dim * np.ones((2, dim))
    centers[0] *= -1
    X = separability * centers[y] + noise
    return X, y


###############################################################################
# Code to run the cross-validations