pretty much the version running on ec2

This commit is contained in:
Nicolo Fusi 2013-02-06 11:43:23 +00:00
parent 6a6cbb58c8
commit 6959751149

View file

@ -123,13 +123,15 @@ class opt_SGD(Optimizer):
else:
raise NotImplementedError
def step_with_missing_data(self, f_fp, X, Y, step, shapes):
def step_with_missing_data(self, f_fp, X, step, shapes):
N, Q = X.shape
samples = self.non_null_samples(self.model.Y)
j = self.subset_parameter_vector(self.x_opt, samples, shapes)
self.model.N = samples.sum()
self.model.X = X[samples]
self.model.Y = self.model.Y[samples]
# self.model.Y -= self.model.Y.mean() # <----------------- WARNING!!!!
# self.model.Y /= self.model.Y.std()
model_name = self.model.__class__.__name__
if model_name == 'Bayesian_GPLVM':
@ -142,11 +144,7 @@ class opt_SGD(Optimizer):
momentum_term = self.momentum * step[j]
try:
f, fp = f_fp(self.x_opt[j])
except Exception:
return 0, step, self.model.N
f, fp = f_fp(self.x_opt[j])
step[j] = self.learning_rate[j] * fp
self.x_opt[j] -= step[j] + momentum_term
@ -183,8 +181,10 @@ class opt_SGD(Optimizer):
self.model.Y = Y[:, j]
# self.model.trYYT = np.sum(np.square(self.model.Y))
if missing_data:
if self.model.Y.std() == 0.0 or self.model.Y.shape[0] == 0:
continue
shapes = self.get_param_shapes(N, Q)
f, step, Nj = self.step_with_missing_data(f_fp, X, Y, step, shapes)
f, step, Nj = self.step_with_missing_data(f_fp, X, step, shapes)
else:
Nj = N
momentum_term = self.momentum * step # compute momentum using update(t-1)