def get_discriminator(self):
''' specify discriminator D0 '''
"""
disc0_layers = [LL.InputLayer(shape=(self.args.batch_size, 3, 32, 32))]
disc0_layers.append(LL.GaussianNoiseLayer(disc0_layers[-1], sigma=0.05))
disc0_layers.append(dnn.Conv2DDNNLayer(disc0_layers[-1], 96, (3,3), pad=1, W=Normal(0.02), nonlinearity=nn.lrelu))
disc0_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc0_layers[-1], 96, (3,3), pad=1, stride=2, W=Normal(0.02), nonlinearity=nn.lrelu))) # 16x16
disc0_layers.append(LL.DropoutLayer(disc0_layers[-1], p=0.1))
disc0_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc0_layers[-1], 192, (3,3), pad=1, W=Normal(0.02), nonlinearity=nn.lrelu)))
disc0_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc0_layers[-1], 192, (3,3), pad=1, stride=2, W=Normal(0.02), nonlinearity=nn.lrelu))) # 8x8
disc0_layers.append(LL.DropoutLayer(disc0_layers[-1], p=0.1))
disc0_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc0_layers[-1], 192, (3,3), pad=0, W=Normal(0.02), nonlinearity=nn.lrelu))) # 6x6
disc0_layer_shared = LL.NINLayer(disc0_layers[-1], num_units=192, W=Normal(0.02), nonlinearity=nn.lrelu) # 6x6
disc0_layers.append(disc0_layer_shared)
disc0_layer_z_recon = LL.DenseLayer(disc0_layer_shared, num_units=50, W=Normal(0.02), nonlinearity=None)
disc0_layers.append(disc0_layer_z_recon) # also need to recover z from x
disc0_layers.append(LL.GlobalPoolLayer(disc0_layer_shared))
disc0_layer_adv = LL.DenseLayer(disc0_layers[-1], num_units=10, W=Normal(0.02), nonlinearity=None)
disc0_layers.append(disc0_layer_adv)
return disc0_layers, disc0_layer_adv, disc0_layer_z_recon
"""
disc_x_layers = [LL.InputLayer(shape=(None, 3, 32, 32))]
disc_x_layers.append(LL.GaussianNoiseLayer(disc_x_layers[-1], sigma=0.2))
disc_x_layers.append(dnn.Conv2DDNNLayer(disc_x_layers[-1], 96, (3,3), pad=1, W=Normal(0.01), nonlinearity=nn.lrelu))
disc_x_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc_x_layers[-1], 96, (3,3), pad=1, stride=2, W=Normal(0.01), nonlinearity=nn.lrelu)))
disc_x_layers.append(LL.DropoutLayer(disc_x_layers[-1], p=0.5))
disc_x_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc_x_layers[-1], 192, (3,3), pad=1, W=Normal(0.01), nonlinearity=nn.lrelu)))
disc_x_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc_x_layers[-1], 192, (3,3), pad=1, stride=2, W=Normal(0.01), nonlinearity=nn.lrelu)))
disc_x_layers.append(LL.DropoutLayer(disc_x_layers[-1], p=0.5))
disc_x_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc_x_layers[-1], 192, (3,3), pad=0, W=Normal(0.01), nonlinearity=nn.lrelu)))
disc_x_layers_shared = LL.NINLayer(disc_x_layers[-1], num_units=192, W=Normal(0.01), nonlinearity=nn.lrelu)
disc_x_layers.append(disc_x_layers_shared)
disc_x_layer_z_recon = LL.DenseLayer(disc_x_layers_shared, num_units=self.args.z0dim, nonlinearity=None)
disc_x_layers.append(disc_x_layer_z_recon) # also need to recover z from x
# disc_x_layers.append(nn.MinibatchLayer(disc_x_layers_shared, num_kernels=100))
disc_x_layers.append(LL.GlobalPoolLayer(disc_x_layers_shared))
disc_x_layer_adv = LL.DenseLayer(disc_x_layers[-1], num_units=10, W=Normal(0.01), nonlinearity=None)
disc_x_layers.append(disc_x_layer_adv)
#output_before_softmax_x = LL.get_output(disc_x_layer_adv, x, deterministic=False)
#output_before_softmax_gen = LL.get_output(disc_x_layer_adv, gen_x, deterministic=False)
# temp = LL.get_output(gen_x_layers[-1], deterministic=False, init=True)
# temp = LL.get_output(disc_x_layers[-1], x, deterministic=False, init=True)
# init_updates = [u for l in LL.get_all_layers(gen_x_layers)+LL.get_all_layers(disc_x_layers) for u in getattr(l,'init_updates',[])]
return disc_x_layers, disc_x_layer_adv, disc_x_layer_z_recon
def build_1Dregression_v1(input_var=None,
input_width=None,
nin_units=12,
h_num_units=[64, 64],
h_grad_clip=1.0,
output_width=1):
"""
A stacked bidirectional RNN network for regression, alternating
with dense layers and merging of the two directions, followed by
a feature mean pooling in the time direction, with a linear
dim-reduction layer at the start
Args:
input_var (theano 3-tensor): minibatch of input sequence vectors
input_width (int): length of input sequences
nin_units (list): number of NIN features
h_num_units (int list): no. of units in hidden layer in each stack
from bottom to top
h_grad_clip (float): gradient clipping maximum value
output_width (int): size of output layer (e.g. =1 for 1D regression)
Returns:
output layer (Lasagne layer object)
"""
# Non-linearity hyperparameter
nonlin = lasagne.nonlinearities.LeakyRectify(leakiness=0.15)
# Input layer
l_in = LL.InputLayer(shape=(None, 22, input_width), input_var=input_var)
batchsize = l_in.input_var.shape[0]
# NIN-layer
l_in = LL.NINLayer(l_in,
num_units=nin_units,
nonlinearity=lasagne.nonlinearities.linear)
l_in_1 = LL.DimshuffleLayer(l_in, (0, 2, 1))
# RNN layers
for h in h_num_units:
# Forward layers
l_forward_0 = LL.RecurrentLayer(l_in_1,
nonlinearity=nonlin,
num_units=h,
backwards=False,
learn_init=True,
grad_clipping=h_grad_clip,
unroll_scan=True,
precompute_input=True)
l_forward_0a = LL.ReshapeLayer(l_forward_0, (-1, h))
l_forward_0b = LL.DenseLayer(l_forward_0a,
num_units=h,
nonlinearity=nonlin)
l_forward_0c = LL.ReshapeLayer(l_forward_0b,
(batchsize, input_width, h))
# Backward layers
l_backward_0 = LL.RecurrentLayer(l_in_1,
nonlinearity=nonlin,
num_units=h,
backwards=True,
learn_init=True,
grad_clipping=h_grad_clip,
unroll_scan=True,
precompute_input=True)
l_backward_0a = LL.ReshapeLayer(l_backward_0, (-1, h))
l_backward_0b = LL.DenseLayer(l_backward_0a,
num_units=h,
nonlinearity=nonlin)
l_backward_0c = LL.ReshapeLayer(l_backward_0b,
(batchsize, input_width, h))
l_in_1 = LL.ElemwiseSumLayer([l_forward_0c, l_backward_0c])
# Output layers
network_0a = LL.ReshapeLayer(l_in_1, (-1, h_num_units[-1]))
network_0b = LL.DenseLayer(network_0a,
num_units=output_width,
nonlinearity=nonlin)
network_0c = LL.ReshapeLayer(network_0b,
(batchsize, input_width, output_width))
output_net_1 = LL.FlattenLayer(network_0c, outdim=2)
output_net_2 = LL.FeaturePoolLayer(output_net_1,
pool_size=input_width,
pool_function=T.mean)
return output_net_2
nonlinearity=nn.lrelu,
name='gz2'),
g=None))
genz_layers.append(
nn.batch_norm(dnn.Conv2DDNNLayer(genz_layers[-1],
512, (3, 3),
pad=1,
stride=2,
W=Normal(0.05),
nonlinearity=nn.lrelu,
name='gz2'),
g=None))
genz_layers.append(
nn.batch_norm(ll.NINLayer(genz_layers[-1],
num_units=512,
W=Normal(0.05),
nonlinearity=nn.lrelu,
name='gz5'),
g=None))
genz_layers.append(ll.GlobalPoolLayer(genz_layers[-1], name='gz6'))
genz_layers.append(
ll.DenseLayer(genz_layers[-1],
num_units=100,
W=Normal(0.05),
nonlinearity=lasagne.nonlinearities.sigmoid,
name='gz7'))
# specify discriminative model
# for z
discz_layers = [z_input]
gen_layers.append(nn.weight_norm(nn.Deconv2DLayer(gen_layers[-1], (args.batch_size,3,32,32), (5,5), W=Normal(0.05), nonlinearity=T.tanh), train_g=True, init_stdv=0.1)) # 16 -> 32 gen_dat = ll.get_output(gen_layers[-1]) # specify discriminative model disc_layers = [ll.InputLayer(shape=(None, 3, 32, 32))] disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.2)) disc_layers.append(nn.weight_norm(dnn.Conv2DDNNLayer(disc_layers[-1], 96, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(nn.weight_norm(dnn.Conv2DDNNLayer(disc_layers[-1], 96, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(nn.weight_norm(dnn.Conv2DDNNLayer(disc_layers[-1], 96, (3,3), pad=1, stride=2, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.5)) disc_layers.append(nn.weight_norm(dnn.Conv2DDNNLayer(disc_layers[-1], 192, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(nn.weight_norm(dnn.Conv2DDNNLayer(disc_layers[-1], 192, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(nn.weight_norm(dnn.Conv2DDNNLayer(disc_layers[-1], 192, (3,3), pad=1, stride=2, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.5)) disc_layers.append(nn.weight_norm(dnn.Conv2DDNNLayer(disc_layers[-1], 192, (3,3), pad=0, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(nn.weight_norm(ll.NINLayer(disc_layers[-1], num_units=192, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(nn.weight_norm(ll.NINLayer(disc_layers[-1], num_units=192, W=Normal(0.05), nonlinearity=nn.lrelu))) disc_layers.append(ll.GlobalPoolLayer(disc_layers[-1])) disc_layers.append(nn.weight_norm(ll.DenseLayer(disc_layers[-1], num_units=16, W=Normal(0.05), nonlinearity=None), train_g=True, init_stdv=0.1)) disc_params = ll.get_all_params(disc_layers, trainable=True) x_temp = T.tensor4() temp = ll.get_output(gen_layers[-1], deterministic=False, init=True) temp = ll.get_output(disc_layers[-1], x_temp, deterministic=False, init=True) init_updates = [u for l in gen_layers+disc_layers for u in getattr(l,'init_updates',[])] init_param = th.function(inputs=[x_temp], outputs=None, updates=init_updates) # costs labels = T.ivector()
# classifier x2y: p_c(x, y) = p(x) p_c(y | x) cla_in_x = ll.InputLayer(shape=(None, in_channels) + dim_input) cla_layers = [cla_in_x] cla_layers.append(ll.DropoutLayer(cla_layers[-1], p=0.2, name='cla-00')) cla_layers.append(ll.batch_norm(dnn.Conv2DDNNLayer(cla_layers[-1], 128, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-02'), name='cla-03')) cla_layers.append(ll.batch_norm(dnn.Conv2DDNNLayer(cla_layers[-1], 128, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-11'), name='cla-12')) cla_layers.append(ll.batch_norm(dnn.Conv2DDNNLayer(cla_layers[-1], 128, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-21'), name='cla-22')) cla_layers.append(dnn.MaxPool2DDNNLayer(cla_layers[-1], pool_size=(2, 2))) cla_layers.append(ll.DropoutLayer(cla_layers[-1], p=0.5, name='cla-23')) cla_layers.append(ll.batch_norm(dnn.Conv2DDNNLayer(cla_layers[-1], 256, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-31'), name='cla-32')) cla_layers.append(ll.batch_norm(dnn.Conv2DDNNLayer(cla_layers[-1], 256, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-41'), name='cla-42')) cla_layers.append(ll.batch_norm(dnn.Conv2DDNNLayer(cla_layers[-1], 256, (3,3), pad=1, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-51'), name='cla-52')) cla_layers.append(dnn.MaxPool2DDNNLayer(cla_layers[-1], pool_size=(2, 2))) cla_layers.append(ll.DropoutLayer(cla_layers[-1], p=0.5, name='cla-53')) cla_layers.append(ll.batch_norm(dnn.Conv2DDNNLayer(cla_layers[-1], 512, (3,3), pad=0, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-61'), name='cla-62')) cla_layers.append(ll.batch_norm(ll.NINLayer(cla_layers[-1], num_units=256, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-71'), name='cla-72')) cla_layers.append(ll.batch_norm(ll.NINLayer(cla_layers[-1], num_units=128, W=Normal(0.05), nonlinearity=nn.lrelu, name='cla-81'), name='cla-82')) cla_layers.append(ll.GlobalPoolLayer(cla_layers[-1], name='cla-83')) cla_layers.append(ll.batch_norm(ll.DenseLayer(cla_layers[-1], num_units=num_classes, W=Normal(0.05), nonlinearity=ln.softmax, name='cla-91'), name='cla-92')) # generator y2x: p_g(x, y) = p(y) p_g(x | y) where x = G(z, y), z follows p_g(z) gen_in_z = ll.InputLayer(shape=(None, n_z)) gen_in_y = ll.InputLayer(shape=(None,)) gen_layers = [gen_in_z] gen_layers.append(MLPConcatLayer([gen_layers[-1], gen_in_y], num_classes, name='gen-00')) gen_layers.append(nn.batch_norm(ll.DenseLayer(gen_layers[-1], num_units=4*4*512, W=Normal(0.05), nonlinearity=nn.relu, name='gen-01'), g=None, name='gen-02')) gen_layers.append(ll.ReshapeLayer(gen_layers[-1], (-1,512,4,4), name='gen-03')) gen_layers.append(ConvConcatLayer([gen_layers[-1], gen_in_y], num_classes, name='gen-10')) gen_layers.append(nn.batch_norm(nn.Deconv2DLayer(gen_layers[-1], (None,256,8,8), (5,5), W=Normal(0.05), nonlinearity=nn.relu, name='gen-11'), g=None, name='gen-12')) # 4 -> 8 gen_layers.append(ConvConcatLayer([gen_layers[-1], gen_in_y], num_classes, name='gen-20')) gen_layers.append(nn.batch_norm(nn.Deconv2DLayer(gen_layers[-1], (None,128,16,16), (5,5), W=Normal(0.05), nonlinearity=nn.relu, name='gen-21'), g=None, name='gen-22')) # 8 -> 16
dnn.Conv2DDNNLayer(disc0_layers[-1],
192, (3, 3),
pad=1,
stride=2,
W=Normal(0.02),
nonlinearity=nn.lrelu))) # 8x8
disc0_layers.append(LL.DropoutLayer(disc0_layers[-1], p=0.1))
disc0_layers.append(
nn.batch_norm(
dnn.Conv2DDNNLayer(disc0_layers[-1],
192, (3, 3),
pad=0,
W=Normal(0.02),
nonlinearity=nn.lrelu))) # 6x6
disc0_layer_shared = LL.NINLayer(disc0_layers[-1],
num_units=192,
W=Normal(0.02),
nonlinearity=nn.lrelu) # 6x6
disc0_layers.append(disc0_layer_shared)
disc0_layer_z_recon = LL.DenseLayer(disc0_layer_shared,
num_units=16,
W=Normal(0.02),
nonlinearity=None)
disc0_layers.append(disc0_layer_z_recon) # also need to recover z from x
disc0_layers.append(LL.GlobalPoolLayer(disc0_layer_shared))
disc0_layer_adv = LL.DenseLayer(disc0_layers[-1],
num_units=10,
W=Normal(0.02),
nonlinearity=None)
disc0_layers.append(disc0_layer_adv)
def get_discriminator_binary():
disc_layers = [ll.InputLayer(shape=(None, 3, 32, 32))]
disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.2))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
96, (3, 3),
pad=1,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
96, (3, 3),
pad=1,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
96, (3, 3),
pad=1,
stride=2,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.5))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
192, (3, 3),
pad=1,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
192, (3, 3),
pad=1,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
192, (3, 3),
pad=1,
stride=2,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.5))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
192, (3, 3),
pad=0,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
ll.NINLayer(disc_layers[-1],
num_units=192,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
ll.NINLayer(disc_layers[-1],
num_units=192,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.GlobalPoolLayer(disc_layers[-1]))
disc_layers.append(
nn.weight_norm(ll.DenseLayer(disc_layers[-1],
num_units=2,
W=Normal(0.05),
nonlinearity=None),
train_g=True,
init_stdv=0.1))
disc_layers.append(
ll.DenseLayer(disc_layers[-2],
num_units=128,
W=Normal(0.05),
nonlinearity=T.nnet.sigmoid))
return disc_layers
pad=1,
stride=2,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.5))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
512, (3, 3),
pad=0,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
ll.NINLayer(disc_layers[-1],
num_units=256,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
ll.NINLayer(disc_layers[-1],
num_units=128,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.GlobalPoolLayer(disc_layers[-1]))
disc_layers.append(
nn.weight_norm(ll.DenseLayer(disc_layers[-1],
num_units=10,
W=Normal(0.05),
nonlinearity=None),
train_g=True,
init_stdv=0.1))
stride=2,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.5))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
disc_dim * 4, (3, 3),
pad=0,
stride=2,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
ll.NINLayer(disc_layers[-1],
num_units=disc_dim * 4,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.GlobalPoolLayer(disc_layers[-1]))
disc_layers.append(nn.MinibatchLayer(disc_layers[-1], num_kernels=50))
disc_layers.append(
nn.weight_norm(ll.DenseLayer(disc_layers[-1],
num_units=1,
W=Normal(0.05),
nonlinearity=None),
train_g=True,
init_stdv=0.1))
# costs
x_lab = T.tensor4()
temp = ll.get_output(gen_layers[-1], deterministic=False, init=True)
temp = ll.get_output(disc_layers[-1], x_lab, deterministic=False, init=True)
# disc_x_layers.append(LL.GlobalPoolLayer(disc_x_layers[-1],)) # #disc_x_layers.append(nn.MinibatchLayer(disc_x_layers[-1], num_kernels=100)) # disc_x_layer_adv = LL.DenseLayer(disc_x_layers[-1], num_units=10, W=Normal(0.01), nonlinearity=None) # disc_x_layers.append(disc_x_layer_adv) # specify discriminative model disc_x_layers = [LL.InputLayer(shape=(None, 3, 32, 32))] disc_x_layers.append(LL.GaussianNoiseLayer(disc_x_layers[-1], sigma=0.2)) disc_x_layers.append(dnn.Conv2DDNNLayer(disc_x_layers[-1], 96, (3,3), pad=1, W=Normal(0.01), nonlinearity=nn.lrelu)) disc_x_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc_x_layers[-1], 96, (3,3), pad=1, stride=2, W=Normal(0.01), nonlinearity=nn.lrelu))) disc_x_layers.append(LL.DropoutLayer(disc_x_layers[-1], p=0.5)) disc_x_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc_x_layers[-1], 192, (3,3), pad=1, W=Normal(0.01), nonlinearity=nn.lrelu))) disc_x_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc_x_layers[-1], 192, (3,3), pad=1, stride=2, W=Normal(0.01), nonlinearity=nn.lrelu))) disc_x_layers.append(LL.DropoutLayer(disc_x_layers[-1], p=0.5)) disc_x_layers.append(nn.batch_norm(dnn.Conv2DDNNLayer(disc_x_layers[-1], 192, (3,3), pad=0, W=Normal(0.01), nonlinearity=nn.lrelu))) disc_x_layers_shared = LL.NINLayer(disc_x_layers[-1], num_units=192, W=Normal(0.01), nonlinearity=nn.lrelu) disc_x_layers.append(disc_x_layers_shared) disc_x_layer_z_recon = LL.DenseLayer(disc_x_layers_shared, num_units=50, nonlinearity=None) disc_x_layers.append(disc_x_layer_z_recon) # also need to recover z from x # disc_x_layers.append(nn.MinibatchLayer(disc_x_layers_shared, num_kernels=100)) disc_x_layers.append(LL.GlobalPoolLayer(disc_x_layers_shared)) disc_x_layer_adv = LL.DenseLayer(disc_x_layers[-1], num_units=10, W=Normal(0.01), nonlinearity=None) disc_x_layers.append(disc_x_layer_adv) #output_before_softmax_x = LL.get_output(disc_x_layer_adv, x, deterministic=False) #output_before_softmax_gen = LL.get_output(disc_x_layer_adv, gen_x, deterministic=False) # temp = LL.get_output(gen_x_layers[-1], deterministic=False, init=True) # temp = LL.get_output(disc_x_layers[-1], x, deterministic=False, init=True)
def get_discriminator_brown(num_feature=256):
disc_layers = [ll.InputLayer(shape=(None, 3, 32, 32))]
disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.2))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
96, (3, 3),
pad=1,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
96, (3, 3),
pad=1,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
96, (3, 3),
pad=1,
stride=2,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.5))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
128, (3, 3),
pad=1,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
128, (3, 3),
pad=1,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
128, (3, 3),
pad=1,
stride=2,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.DropoutLayer(disc_layers[-1], p=0.5))
disc_layers.append(
nn.weight_norm(
dnn.Conv2DDNNLayer(disc_layers[-1],
128, (3, 3),
pad=0,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
ll.NINLayer(disc_layers[-1],
num_units=num_feature,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(
nn.weight_norm(
ll.NINLayer(disc_layers[-1],
num_units=128,
W=Normal(0.05),
nonlinearity=nn.lrelu)))
disc_layers.append(ll.GlobalPoolLayer(disc_layers[-1]))
disc_layers.append(
nn.weight_norm(ll.DenseLayer(disc_layers[-1],
num_units=2,
W=Normal(0.05),
nonlinearity=None),
train_g=True,
init_stdv=0.1))
#disc_layers.append(ll.ReshapeLayer(disc_layers[-4], ([0], -1)))
#disc_layers.append(ll.GlobalPoolLayer(disc_layers[-4]))
disc_layer_features_low_dim = -4
disc_layer_features_high_dim = -5
return disc_layers, disc_layer_features_low_dim, disc_layer_features_high_dim
layers.append(
batch_norm(
dnn.Conv2DDNNLayer(layers[-1], 192, (3, 3), pad=1, nonlinearity=f)))
layers.append(
batch_norm(
dnn.Conv2DDNNLayer(layers[-1], 192, (3, 3), pad=1, nonlinearity=f)))
layers.append(
batch_norm(
dnn.Conv2DDNNLayer(layers[-1], 192, (3, 3), pad=1, nonlinearity=f)))
layers.append(ll.MaxPool2DLayer(layers[-1], 2))
layers.append(ll.DropoutLayer(layers[-1], p=0.5))
layers.append(
batch_norm(
dnn.Conv2DDNNLayer(layers[-1], 192, (3, 3), pad=0, nonlinearity=f)))
layers.append(
batch_norm(ll.NINLayer(layers[-1], num_units=192, nonlinearity=f)))
layers.append(
batch_norm(ll.NINLayer(layers[-1], num_units=192, nonlinearity=f)))
layers.append(nn.GlobalAvgLayer(layers[-1]))
layers.append(
batch_norm(ll.DenseLayer(layers[-1], num_units=10, nonlinearity=None)))
# discriminative cost & updates
output_before_softmax = ll.get_output(layers[-1], x)
y = T.ivector()
cost = nn.softmax_loss(y, output_before_softmax)
train_err = T.mean(T.neq(T.argmax(output_before_softmax, axis=1), y))
params = ll.get_all_params(layers, trainable=True)
lr = T.scalar()
mom1 = T.scalar()
param_updates = nn.adam_updates(params, cost, lr=lr, mom1=mom1)
raise NotImplementedError('incorrect norm type')
layers = [ll.InputLayer(shape=(None, 3, 32, 32))]
layers.append(ll.GaussianNoiseLayer(layers[-1], sigma=0.15))
layers.append(normalizer(dnn.Conv2DDNNLayer(layers[-1], 96, (3,3), pad=1, nonlinearity=nn.lrelu)))
layers.append(normalizer(dnn.Conv2DDNNLayer(layers[-1], 96, (3,3), pad=1, nonlinearity=nn.lrelu)))
layers.append(normalizer(dnn.Conv2DDNNLayer(layers[-1], 96, (3,3), pad=1, nonlinearity=nn.lrelu)))
layers.append(ll.MaxPool2DLayer(layers[-1], 2))
layers.append(ll.DropoutLayer(layers[-1], p=0.5))
layers.append(normalizer(dnn.Conv2DDNNLayer(layers[-1], 192, (3,3), pad=1, nonlinearity=nn.lrelu)))
layers.append(normalizer(dnn.Conv2DDNNLayer(layers[-1], 192, (3,3), pad=1, nonlinearity=nn.lrelu)))
layers.append(normalizer(dnn.Conv2DDNNLayer(layers[-1], 192, (3,3), pad=1, nonlinearity=nn.lrelu)))
layers.append(ll.MaxPool2DLayer(layers[-1], 2))
layers.append(ll.DropoutLayer(layers[-1], p=0.5))
layers.append(normalizer(dnn.Conv2DDNNLayer(layers[-1], 192, (3,3), pad=0, nonlinearity=nn.lrelu)))
layers.append(normalizer(ll.NINLayer(layers[-1], num_units=192, nonlinearity=nn.lrelu)))
layers.append(normalizer(ll.NINLayer(layers[-1], num_units=192, nonlinearity=nn.lrelu)))
layers.append(nn.GlobalAvgLayer(layers[-1]))
layers.append(normalizer(ll.DenseLayer(layers[-1], num_units=10, nonlinearity=None)))
# initialization
x = T.tensor4()
temp = ll.get_output(layers[-1], x, init=True)
init_updates = [u for l in layers for u in getattr(l,'init_updates',[])]
# discriminative cost & updates
output_before_softmax = ll.get_output(layers[-1], x)
bn_updates = [u for l in layers for u in getattr(l,'bn_updates',[])]
y = T.ivector()
cost = nn.softmax_loss(y, output_before_softmax)
train_err = T.mean(T.neq(T.argmax(output_before_softmax,axis=1),y))
