def fit(self, X):
n_samples, n_feats = X.shape
n_train_batches = n_samples / self.batch_size
v_X = share_data(X)
self.da_ = DenoisingAutoEncoderFormula(n_visible = n_feats,
n_hidden = self.n_hidden,
corruption_level = self.corruption_level)
index = T.lscalar('index')
cost = self.da_.cost()
gparams = T.grad(cost, self.da_.params)
updates = [(param, param - self.learning_rate*gparam)
for (param, gparam) in zip(self.da_.params, gparams)]
train_model = theano.function(inputs = [index],
outputs = cost,
updates = updates,
givens = {
self.da_.X: v_X[index*self.batch_size:(index+1)*self.batch_size]
})
for epoch in xrange(self.n_epochs):
cost = [train_model(i) for i in xrange(n_train_batches)]
if self.verbose:
print 'training epoch %d, recall cost %f ' % (
epoch, np.mean(cost))
return self
def fit(self, X):
n_samples, n_feats = X.shape
n_train_batches = n_samples / self.batch_size
v_X = share_data(X)
self.ca_ = ContractiveAutoEncoderFormula(n_visible = n_feats,
n_hidden = self.n_hidden,
batch_size = self.batch_size,
contraction_level = self.contraction_level)
index = T.lscalar('index')
cost = self.ca_.cost()
gparams = T.grad(cost, self.ca_.params)
updates = [(param, param - self.learning_rate*gparam)
for (param, gparam) in zip(self.ca_.params, gparams)]
train_model = theano.function(inputs = [index],
outputs = [T.mean(self.ca_.L_rec), self.ca_.L_jacob],
updates = updates,
givens = {
self.ca_.X: v_X[index*self.batch_size:(index+1)*self.batch_size]
})
for epoch in xrange(self.n_epochs):
recall, jacob = zip(*[train_model(i) for i in xrange(n_train_batches)])
if self.verbose:
print 'training epoch %d, recall cost %f, jacobian norm %f ' % (
epoch, np.mean(recall), np.mean(np.sqrt(jacob))
)
return self
def fit(self, X):
n_samples, n_feats = X.shape
n_train_batches = n_samples / self.batch_size
v_X = share_data(X)
self.ca_ = ContractiveAutoEncoderFormula(
n_visible=n_feats,
n_hidden=self.n_hidden,
batch_size=self.batch_size,
contraction_level=self.contraction_level)
index = T.lscalar('index')
cost = self.ca_.cost()
gparams = T.grad(cost, self.ca_.params)
updates = [(param, param - self.learning_rate * gparam)
for (param, gparam) in zip(self.ca_.params, gparams)]
train_model = theano.function(
inputs=[index],
outputs=[T.mean(self.ca_.L_rec), self.ca_.L_jacob],
updates=updates,
givens={
self.ca_.X:
v_X[index * self.batch_size:(index + 1) * self.batch_size]
})
for epoch in xrange(self.n_epochs):
recall, jacob = zip(
*[train_model(i) for i in xrange(n_train_batches)])
if self.verbose:
print 'training epoch %d, recall cost %f, jacobian norm %f ' % (
epoch, np.mean(recall), np.mean(np.sqrt(jacob)))
return self
def fit(self, X):
n_samples, n_feats = X.shape
n_train_batches = n_samples / self.batch_size
v_X = share_data(X)
self.da_ = DenoisingAutoEncoderFormula(
n_visible=n_feats,
n_hidden=self.n_hidden,
corruption_level=self.corruption_level)
index = T.lscalar('index')
cost = self.da_.cost()
gparams = T.grad(cost, self.da_.params)
updates = [(param, param - self.learning_rate * gparam)
for (param, gparam) in zip(self.da_.params, gparams)]
train_model = theano.function(
inputs=[index],
outputs=cost,
updates=updates,
givens={
self.da_.X:
v_X[index * self.batch_size:(index + 1) * self.batch_size]
})
for epoch in xrange(self.n_epochs):
cost = [train_model(i) for i in xrange(n_train_batches)]
if self.verbose:
print 'training epoch %d, recall cost %f ' % (epoch,
np.mean(cost))
return self
def transform(self, X): v_X = share_data(X) return self.ca_.hidden_value(v_X).eval()
def transform(self, X):
v_X = share_data(X)
return self.ca_.hidden_value(v_X).eval()
