def test_model( n_in
,n_feat
,n_hidden
,n_out
,learning_rate=0.01
, n_epochs=1000
, dataset = '../deep_training/new_model_train6'
, batch_size = 20):
train_set_x, train_set_y, train_set_z = load_data(dataset, n_out)
n_train_batches = train_set_x.get_value(borrow=True).shape[0] / batch_size
print '... building the model'
index = T.lscalar()
x = T.matrix('x')
z = T.matrix('z')
classifier = new_model(input=x, n_in=24, n_feat=5, n_hidden = 50, n_out=n_out)
cost = classifier.mean_square_error(z)
gparams = []
for param in classifier.params:
gparam = T.grad(cost, param)
gparams.append(gparam)
# specify how to update the parameters of the model as a dictionary
updates = {}
for param, gparam in zip(classifier.params, gparams):
updates[param] = param - learning_rate*gparam
train_model =theano.function( inputs = [index], outputs = cost,
updates = updates,
givens={
x:train_set_x[index*batch_size:(index+1)*batch_size],
z:train_set_z[index*batch_size:(index+1)*batch_size]})
execute_model = theano.function(inputs = [index],
outputs = classifier.p_y_given_x,
givens={x:train_set_x[index*batch_size:(index+1)*batch_size]})
print '... training'
start_time = time.clock()
done_looping = False
epoch = 0
while (epoch < n_epochs) and (not done_looping):
epoch = epoch + 1
shuffled_idx = range(n_train_batches)
numpy.random.shuffle(shuffled_idx)
for minibatch_index in shuffled_idx:
minibatch_avg_cost = train_model(minibatch_index)
test_distrib = [execute_model(i) for i in xrange(n_train_batches)]
test_distrib = numpy.vstack(test_distrib)
approx = [t for t in test_distrib]
avg_sim = numpy.mean(similarity(distrib, approx))
print "Similarity:", avg_sim
end_time = time.clock()
print "... done in", end_time - start_time
def test_model(n_in,
n_feat,
n_hidden,
n_out,
learning_rate=0.01,
n_epochs=1000,
dataset='../deep_training/new_model_train6',
batch_size=20):
train_set_x, train_set_y, train_set_z = load_data(dataset, n_out)
n_train_batches = train_set_x.get_value(borrow=True).shape[0] / batch_size
print '... building the model'
index = T.lscalar()
x = T.matrix('x')
z = T.matrix('z')
classifier = new_model(input=x,
n_in=24,
n_feat=5,
n_hidden=50,
n_out=n_out)
cost = classifier.mean_square_error(z)
gparams = []
for param in classifier.params:
gparam = T.grad(cost, param)
gparams.append(gparam)
# specify how to update the parameters of the model as a dictionary
updates = {}
for param, gparam in zip(classifier.params, gparams):
updates[param] = param - learning_rate * gparam
train_model = theano.function(
inputs=[index],
outputs=cost,
updates=updates,
givens={
x: train_set_x[index * batch_size:(index + 1) * batch_size],
z: train_set_z[index * batch_size:(index + 1) * batch_size]
})
execute_model = theano.function(
inputs=[index],
outputs=classifier.p_y_given_x,
givens={x: train_set_x[index * batch_size:(index + 1) * batch_size]})
print '... training'
start_time = time.clock()
done_looping = False
epoch = 0
while (epoch < n_epochs) and (not done_looping):
epoch = epoch + 1
shuffled_idx = range(n_train_batches)
numpy.random.shuffle(shuffled_idx)
for minibatch_index in shuffled_idx:
minibatch_avg_cost = train_model(minibatch_index)
test_distrib = [execute_model(i) for i in xrange(n_train_batches)]
test_distrib = numpy.vstack(test_distrib)
approx = [t for t in test_distrib]
avg_sim = numpy.mean(similarity(distrib, approx))
print "Similarity:", avg_sim
end_time = time.clock()
print "... done in", end_time - start_time
print "... similarity",
f = open('../models/similarity6', 'r')
distrib = pickle.load(f)
esn_distrib = []
for s in sentences:
reservoir.states = np.c_[initial_state].T
for i in range(len(s) - 1):
x = reservoir.execute(np.dot(np.c_[lut[s[i]]].T, features))[0]
x = np.dot(x, readout)
x[np.nonzero(x < 0)] = 0
if sum(x) == 0:
x[0] = 0.0001
x = x / sum(x)
esn_distrib.append(list(x))
print np.mean(similarity(distrib, esn_distrib))
import pylab
pylab.ion()
fig1 = pylab.figure(1)
pylab.draw()
pylab.title('Before')
ax1 = fig1.add_subplot(111, autoscale_on=True)
ax1.plot(features[:, 0], features[:, 1], 'x')
pca = mdp.nodes.PCANode(output_dim=2)
pca.train(features)
pca.stop_training()
reduced_features = pca.execute(features)
for w in vocabulary:
print "... similarity",
f = open("../models/similarity6", "r")
distrib = pickle.load(f)
esn_distrib = []
for s in sentences:
reservoir.states = np.c_[initial_state].T
for i in range(len(s) - 1):
x = reservoir.execute(np.dot(np.c_[lut[s[i]]].T, features))
x = readout.execute(x)[0]
x[np.nonzero(x < 0)] = 0
if sum(x) == 0:
x[0] = 0.0001
x = x / sum(x)
esn_distrib.append(list(x))
print np.mean(similarity(distrib, esn_distrib))
import pylab
pylab.ion()
fig1 = pylab.figure(1)
pylab.draw()
pylab.title("Before")
ax1 = fig1.add_subplot(111, autoscale_on=True)
ax1.plot(features[:, 0], features[:, 1], "x")
pca = mdp.nodes.PCANode(output_dim=2)
pca.train(features)
pca.stop_training()
reduced_features = pca.execute(features)
