def compute_kernel_slow(Y, X, kernel, h):
d = np.sqrt(((Y[:, None, :] - X) ** 2).sum(-1))
norm = kernel_norm(h, X.shape[1], kernel)
if kernel == 'gaussian':
return norm * np.exp(-0.5 * (d * d) / (h * h)).sum(-1)
elif kernel == 'tophat':
return norm * (d < h).sum(-1)
elif kernel == 'epanechnikov':
return norm * ((1.0 - (d * d) / (h * h)) * (d < h)).sum(-1)
elif kernel == 'exponential':
return norm * (np.exp(-d / h)).sum(-1)
elif kernel == 'linear':
return norm * ((1 - d / h) * (d < h)).sum(-1)
elif kernel == 'cosine':
return norm * (np.cos(0.5 * np.pi * d / h) * (d < h)).sum(-1)
else:
raise ValueError('kernel not recognized')
def compute_kernel_slow(Y, X, kernel, h):
d = np.sqrt(((Y[:, None, :] - X) ** 2).sum(-1))
norm = kernel_norm(h, X.shape[1], kernel)
if kernel == 'gaussian':
return norm * np.exp(-0.5 * (d * d) / (h * h)).sum(-1)
elif kernel == 'tophat':
return norm * (d < h).sum(-1)
elif kernel == 'epanechnikov':
return norm * ((1.0 - (d * d) / (h * h)) * (d < h)).sum(-1)
elif kernel == 'exponential':
return norm * (np.exp(-d / h)).sum(-1)
elif kernel == 'linear':
return norm * ((1 - d / h) * (d < h)).sum(-1)
elif kernel == 'cosine':
return norm * (np.cos(0.5 * np.pi * d / h) * (d < h)).sum(-1)
else:
raise ValueError('kernel not recognized')
def compute_kernel_slow(Y, X, kernel, h):
d = np.sqrt(((Y[:, None, :] - X) ** 2).sum(-1))
norm = kernel_norm(h, X.shape[1], kernel)
if kernel == "gaussian":
return norm * np.exp(-0.5 * (d * d) / (h * h)).sum(-1)
elif kernel == "tophat":
return norm * (d < h).sum(-1)
elif kernel == "epanechnikov":
return norm * ((1.0 - (d * d) / (h * h)) * (d < h)).sum(-1)
elif kernel == "exponential":
return norm * (np.exp(-d / h)).sum(-1)
elif kernel == "linear":
return norm * ((1 - d / h) * (d < h)).sum(-1)
elif kernel == "cosine":
return norm * (np.cos(0.5 * np.pi * d / h) * (d < h)).sum(-1)
else:
raise ValueError("kernel not recognized")