def log_likelihoods(self, data, input, mask, tag, x):
etas, nus = np.exp(self.inv_etas), np.exp(self.inv_nus)
mus = self.forward(x, input, tag)
return independent_studentst_logpdf(data[:, None, :],
mus,
etas,
nus,
mask=mask[:, None, :])
def log_likelihoods(self, data, input, mask, tag):
D, mus, sigmas, nus = self.D, self.mus, np.exp(
self.inv_sigmas), np.exp(self.inv_nus)
nus = np.tile(nus[:, None], (1, self.D))
mask = np.ones_like(data, dtype=bool) if mask is None else mask
return stats.independent_studentst_logpdf(data[:, None, :],
mus,
sigmas,
nus,
mask=mask[:, None, :])
def log_likelihoods(self, data, input, mask, tag):
D = self.D
mus = self._compute_mus(data, input, mask, tag)
sigmas = self._compute_sigmas(data, input, mask, tag)
nus = np.exp(self.inv_nus)
nus = np.tile(nus[:, None], (1, self.D))
return stats.independent_studentst_logpdf(data[:, None, :],
mus,
sigmas,
nus,
mask=mask[:, None, :])
def test_independent_studentst_logpdf(T=100, K=4, D=10):
# Test single datapoint log pdf
x = npr.randn(T, D)
mu = npr.randn(K, D)
sigmasqs = np.exp(npr.randn(K, D))
nus = np.exp(npr.randn(K, D))
ll1 = independent_studentst_logpdf(x[:, None, :], mu, sigmasqs, nus)
ll2 = np.sum(t.logpdf(x[:, None, :],
nus[None, :, :],
loc=mu[None, :, :],
scale=np.sqrt(sigmasqs[None, :, :])),
axis=-1)
assert np.allclose(ll1, ll2)