def testStudentsTNllAgainstTfp(self):
"""Check that our Student's T NLL matches TensorFlow Probability."""
for _ in range(10):
x = np.random.normal()
df = np.exp(4. * np.random.normal())
scale = np.exp(4. * np.random.normal())
nll = util.students_t_nll(x, df, scale)
nll_true = -tfp.distributions.StudentT(
df=df, loc=tf.zeros_like(scale), scale=scale).log_prob(x)
self.assertAllClose(nll, nll_true, atol=1e-4, rtol=1e-4)
def lossfun(self, x):
"""A variant of lossfun() that uses the NLL of a Student's t-distribution.
Args:
x: The residual for which the loss is being computed. Must be a rank-2
tensor, where the innermost dimension is the batch index, and the
outermost dimension must be equal to self.num_dims. Must be a tensor or
numpy array of type self.float_dtype.
Returns:
A tensor of the same type and shape as input `x`, containing the loss at
each element of `x`. These "losses" are actually negative log-likelihoods
(as produced by distribution.nllfun()) and so they are not actually
bounded from below by zero. You'll probably want to minimize their sum or
mean.
"""
x = torch.as_tensor(x)
assert len(x.shape) == 2
assert x.shape[1] == self.num_dims
assert x.dtype == self.float_dtype
return util.students_t_nll(x, self.df(), self.scale())