def test_rand_gaussian():
X = np.empty((4000, 1000), dtype=np.float32)
Y = op.rand_gaussian_like(X)
rtol = 1e-3
assert (Y.mean() - 0.0) < rtol, "mean: %f" % Y.mean()
assert Y.std() - 1.0 < rtol, "std: %f" % Y.std()
Y = op.rand_gaussian_like(X, mu=5.0, sigma=2.0)
rtol = 1e-3
assert (Y.mean() - 5.0) < rtol, "mean: %f" % Y.mean()
assert Y.std() - 2.0 < rtol, "std: %f" % Y.std()
Xd = op.to_gpu(X)
Yd = gpuarray.empty_like(Xd)
Y = op.to_cpu(op.rand_gaussian_like(Xd))
rtol = 1e-2
assert (Y.mean() - 0.0) < rtol, "mean: %f" % Y.mean()
assert Y.std() - 1.0 < rtol, "std: %f" % Y.std()
Y = op.to_cpu(op.rand_gaussian_like(Xd, mu=5.0, sigma=2.0))
rtol = 1e-2
assert (Y.mean() - 5.0) < rtol, "mean: %f" % Y.mean()
assert Y.std() - 2.0 < rtol, "std: %f" % Y.std()
def test_rand_gaussian():
X = np.empty((4000, 1000), dtype=np.float32)
Y = op.rand_gaussian_like(X)
rtol = 1e-3
assert (Y.mean() - 0.0) < rtol, "mean: %f" % Y.mean()
assert Y.std() - 1.0 < rtol, "std: %f" % Y.std()
Y = op.rand_gaussian_like(X, mu=5.0, sigma=2.0)
rtol = 1e-3
assert (Y.mean() - 5.0) < rtol, "mean: %f" % Y.mean()
assert Y.std() - 2.0 < rtol, "std: %f" % Y.std()
Xd = op.to_gpu(X)
Yd = gpuarray.empty_like(Xd)
Y = op.to_cpu(op.rand_gaussian_like(Xd))
rtol = 1e-2
assert (Y.mean() - 0.0) < rtol, "mean: %f" % Y.mean()
assert Y.std() - 1.0 < rtol, "std: %f" % Y.std()
Y = op.to_cpu(op.rand_gaussian_like(Xd, mu=5.0, sigma=2.0))
rtol = 1e-2
assert (Y.mean() - 5.0) < rtol, "mean: %f" % Y.mean()
assert Y.std() - 2.0 < rtol, "std: %f" % Y.std()
def _sample_visibles(self, h):
"""Sample from the distribution P(v|h).
Parameters
----------
h : array-like, shape (n_samples, n_components)
Values of the hidden layer to sample from.
Returns
-------
v : array-like, shape (n_samples, n_features)
Values of the visible layer.
"""
p = self._mean_visibles(h)
return p + op.rand_gaussian_like(p)
def _sample_visibles(self, h):
"""Sample from the distribution P(v|h).
Parameters
----------
h : array-like, shape (n_samples, n_components)
Values of the hidden layer to sample from.
Returns
-------
v : array-like, shape (n_samples, n_features)
Values of the visible layer.
"""
p = self._mean_visibles(h)
return p + op.rand_gaussian_like(p)
