In [1]:

    

from pylearn2.models.mlp import MLP, Linear, Tanh, CompositeLayer, Sigmoid


    

    




Using gpu device 0: GeForce GTX 680

In [2]:

    

dimZ = 20
dimX = 784
bsize = 200
IH_model = MLP(batch_size=bsize, nvis=784, layers=[
    Tanh(layer_name='IH', dim=400, irange=.01),
    CompositeLayer(layer_name='mz and vz',
        layers=[
            Linear(layer_name='mz', dim=20, irange=.01),
            Linear(layer_name='vz', dim=20, irange=.01)
        ]
        )
    ]
)

HI_model = MLP(batch_size=bsize, nvis=dimZ, layers=[
    Tanh(layer_name='H_decoder', dim=400, irange=.01),
    Sigmoid(layer_name='x_hat', dim=dimX, irange=.01)
])


    

In [3]:

    

IH_model.get_input_space()


    

    
Out[3]:





VectorSpace(dim=784, dtype=float32)

In [4]:

    

IH_model.get_output_space()


    

    
Out[4]:





CompositeSpace(VectorSpace(dim=20, dtype=float32), VectorSpace(dim=20, dtype=float32))

In [5]:

    

X = IH_model.input_space.make_theano_batch()
Y1, Y2 = IH_model.fprop(X)


    

In [6]:

    

print Y1
print Y2


    

    




mz_z
vz_z

In [7]:

    

IH_model.batch_size


    

    
Out[7]:





200

In [8]:

    

from vpcn import VAE2
import numpy as np
import argparse
import time
import gzip, cPickle

f = gzip.open('../Variational-Autoencoder/mnist.pkl.gz', 'rb')
(x_train, t_train), (x_valid, t_valid), (x_test, t_test)  = cPickle.load(f)
f.close()

data = x_train.astype('float32')

dimZ = 20

batch_size = 200
L = 1
learning_rate = .01

[N,dimX] = data.shape
encoder = VAE2(IH_model,HI_model,L,learning_rate)
encoder.initParams()
encoder.createGradientFunctions()

print "Initializing weights and biases"
lowerbound = np.array([])
testlowerbound = np.array([])

begin = time.time()
for j in xrange(150):
    encoder.lowerbound = 0
    print 'Iteration:', j
    encoder.iterate(data)
    end = time.time()
    print("Iteration %d, lower bound = %.2f,"
          " time = %.2fs"
          % (j, encoder.lowerbound*batch_size/N, end - begin))
    begin = end

    if j % 5 == 0:
        print "Calculating test lowerbound"
        testlowerbound = np.append(testlowerbound,encoder.getLowerBound(x_test.astype('float32')))


    

    




---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-8-d18033893c51> in <module>()
----> 1 from vpcn import VAE2
      2 import numpy as np
      3 import argparse
      4 import time
      5 import gzip, cPickle

ImportError: cannot import name VAE2

In [ ]:

    

%matplotlib inline
from pylab import imshow, rcParams
import display_data 
rcParams['figure.figsize'] = 8, 8
W = IH_model.get_param_values()[0]
I = display_data.displayData(W.T, 'W.png')
imshow(I,cmap='gray')


    

In [ ]:

    

from vpcn import VAE2
import numpy as np
import argparse
import time
import gzip, cPickle

f = gzip.open('../Variational-Autoencoder/mnist.pkl.gz', 'rb')
(x_train, t_train), (x_valid, t_valid), (x_test, t_test)  = cPickle.load(f)
f.close()

data = x_train.astype('float32')

dimZ = 20

batch_size = 100
L = 1
learning_rate = 0.00001

[N,dimX] = data.shape
encoder = VAE2(IH_model,HI_model,L,learning_rate)
encoder.initParams()
encoder.createGradientFunctions()


    

In [ ]: