apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-07-05 16:02:13 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Added some shifts to the degrees of freedom parameter.

Browse source
This commit is contained in:
Joachim van der Herten 2017-07-14 23:48:24 +02:00
parent 5bb17f21d2
commit 394d3ea236
1 changed files with 2 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

6
GPy/models/tp_regression.py
View file

@ -83,9 +83,7 @@ class TPRegression(Model):
self.link_parameter(mean_function) self.link_parameter(mean_function)
# Degrees of freedom # Degrees of freedom
# self.nu = Param('deg_free', float(deg_free), LogexpClipped(lower=2.))
self.nu = Param('deg_free', float(deg_free), Logexp()) self.nu = Param('deg_free', float(deg_free), Logexp())
# self.nu = Param('deg_free', float(deg_free), Logistic(2., np.inf))
self.link_parameter(self.nu) self.link_parameter(self.nu)
# Inference # Inference
@ -245,7 +243,7 @@ class TPRegression(Model):
:rtype: [np.ndarray (Xnew x self.output_dim), np.ndarray (Xnew x self.output_dim)] :rtype: [np.ndarray (Xnew x self.output_dim), np.ndarray (Xnew x self.output_dim)]
""" """
mu, var = self._raw_predict(X, full_cov=False, kern=kern) mu, var = self._raw_predict(X, full_cov=False, kern=kern)
quantiles = [stats.t.ppf(q / 100., self.nu + self.num_data) * np.sqrt(var) + mu for q in quantiles] quantiles = [stats.t.ppf(q / 100., self.nu + 2 + self.num_data) * np.sqrt(var) + mu for q in quantiles]
if self.normalizer is not None: if self.normalizer is not None:
quantiles = [self.normalizer.inverse_mean(q) for q in quantiles] quantiles = [self.normalizer.inverse_mean(q) for q in quantiles]
@ -276,7 +274,7 @@ class TPRegression(Model):
mu, var = self.normalizer.inverse_mean(mu), self.normalizer.inverse_variance(var) mu, var = self.normalizer.inverse_mean(mu), self.normalizer.inverse_variance(var)
def sim_one_dim(m, v): def sim_one_dim(m, v):
nu = self.nu + self.num_data nu = self.nu + 2 + self.num_data
v = np.diag(v.flatten()) if not full_cov else v v = np.diag(v.flatten()) if not full_cov else v
Z = np.random.multivariate_normal(np.zeros(X.shape[0]), v, size).T Z = np.random.multivariate_normal(np.zeros(X.shape[0]), v, size).T
g = np.tile(np.random.gamma(nu / 2., 2. / nu, size), (X.shape[0], 1)) g = np.tile(np.random.gamma(nu / 2., 2. / nu, size), (X.shape[0], 1))