def get_mixture_coef(self, out_tensor):
""" Parses the output tensor to appropriate mixture density coefficients"""
# This uses eqns 18 -> 23 of http://arxiv.org/abs/1308.0850.
# Pen states:
z_pen_logits = out_tensor[:, :, 0:3]
# Process outputs into MDN parameters
M = self.hps['num_mixture']
dist_params = [
out_tensor[:, :, (3 + M * (n - 1)):(3 + M * n)]
for n in range(1, 7)
]
z_pi, z_mu1, z_mu2, z_sigma1, z_sigma2, z_corr = dist_params
# Softmax all the pi's and pen states:
z_pi = softmax(z_pi)
z_pen = softmax(z_pen_logits)
# Exponent the sigmas and also make corr between -1 and 1.
z_sigma1 = exponential(z_sigma1)
z_sigma2 = exponential(z_sigma2)
z_corr = tanh(z_corr)
r = [
z_pi, z_mu1, z_mu2, z_sigma1, z_sigma2, z_corr, z_pen, z_pen_logits
]
return r
def test_exponential(self):
test_values = np.random.random((2, 5))
x = backend.placeholder(ndim=2)
exp = activations.exponential(x)
f = backend.function([x], [exp])
result = f([test_values])[0]
expected = np.exp(test_values)
self.assertAllClose(result, expected, rtol=1e-05)