def __init__(self,
src_path,
tgt_path,
test=False,
src_vocab_size=None,
tgt_vocab_size=None,
sw2i=None,
tw2i=None):
src, tgt = get_dataset(src_path, tgt_path)
if test:
svocab = list(sw2i.keys())
tvocab = list(tw2i.keys())
si2w = {v: k for k, v in sw2i.items()}
ti2w = {v: k for k, v in tw2i.items()}
else:
svocab, sw2i, si2w = build_vocab(utils.flatten(src),
src_vocab_size)
tvocab, tw2i, ti2w = build_vocab(utils.flatten(tgt),
tgt_vocab_size)
# src_p, tgt_p = [], []
# for s, t in zip(src, tgt):
# src_p.append(utils.prepare_sequence(s + ['</s>'],
# sw2i).view(1, -1))
# tgt_p.append(utils.prepare_sequence(t + ['</s>'],
# tw2i).view(1, -1))
self.svocab = svocab
self.tvocab = tvocab
self.sw2i = sw2i
self.tw2i = tw2i
self.si2w = si2w
self.ti2w = ti2w
# self.src = src_p
# self.tgt = tgt_p
self.src = [' '.join(s) for s in src]
self.tgt = [' '.join(t) for t in tgt]
def discriminator(x,
convolutional=True,
filter_sizes=[5, 5, 5, 5],
activation=tf.nn.relu,
n_filters=[100, 100, 100, 100]):
"""Summary
Parameters
----------
x : TYPE
Description
convolutional : bool, optional
Description
filter_sizes : list, optional
Description
activation : TYPE, optional
Description
n_filters : list, optional
Description
Returns
-------
name : TYPE
Description
"""
encoding = encoder(
x=x,
convolutional=convolutional,
dimensions=n_filters,
filter_sizes=filter_sizes,
activation=activation)
# flatten, then linear to 1 value
res = utils.flatten(encoding['z'], name='flatten')
if res.get_shape().as_list()[-1] > 1:
res = utils.linear(res, 1)[0]
return {
'logits': res,
'probs': tf.nn.sigmoid(res),
'Ws': encoding['Ws'],
'hs': encoding['hs']
}
def VAEGAN(input_shape=[None, 784],
n_filters=[64, 64, 64],
filter_sizes=[4, 4, 4],
n_hidden=32,
n_code=2,
activation=tf.nn.tanh,
convolutional=False,
variational=False):
"""Summary
Parameters
----------
input_shape : list, optional
Description
n_filters : list, optional
Description
filter_sizes : list, optional
Description
n_hidden : int, optional
Description
n_code : int, optional
Description
activation : TYPE, optional
Description
convolutional : bool, optional
Description
variational : bool, optional
Description
Returns
-------
name : TYPE
Description
"""
# network input / placeholders for train (bn)
x = tf.placeholder(tf.float32, input_shape, 'x')
z_samp = tf.placeholder(tf.float32, [None, n_code], 'z_samp')
with tf.variable_scope('encoder'):
encoding = encoder(x=x,
n_hidden=n_hidden,
convolutional=convolutional,
dimensions=n_filters,
filter_sizes=filter_sizes,
activation=activation)
with tf.variable_scope('variational'):
z, z_mu, z_log_sigma, loss_z = variational_bayes(h=encoding['z'],
n_code=n_code)
shapes = encoding['shapes'].copy()
shapes.reverse()
n_filters_decoder = n_filters.copy()
n_filters_decoder.reverse()
n_filters_decoder += [input_shape[-1]]
with tf.variable_scope('generator'):
decoding_actual = decoder(z=z,
shapes=shapes,
n_hidden=n_hidden,
convolutional=convolutional,
dimensions=n_filters_decoder,
filter_sizes=filter_sizes,
activation=activation)
with tf.variable_scope('generator', reuse=True):
decoding_sampled = decoder(z=z_samp,
shapes=shapes,
n_hidden=n_hidden,
convolutional=convolutional,
dimensions=n_filters_decoder,
filter_sizes=filter_sizes,
activation=activation)
with tf.variable_scope('discriminator'):
D_real = discriminator(x,
filter_sizes=filter_sizes,
n_filters=n_filters,
activation=activation)
with tf.variable_scope('discriminator', reuse=True):
D_fake = discriminator(decoding_actual['x_tilde'],
filter_sizes=filter_sizes,
n_filters=n_filters,
activation=activation)
with tf.variable_scope('discriminator', reuse=True):
D_samp = discriminator(decoding_sampled['x_tilde'],
filter_sizes=filter_sizes,
n_filters=n_filters,
activation=activation)
with tf.variable_scope('loss'):
# Weights influence of content/style of decoder
gamma = tf.placeholder(tf.float32, name='gamma')
# Discriminator_l Log Likelihood Loss
loss_D_llike = 0
for h_fake, h_real in zip(D_fake['hs'][3:], D_real['hs'][3:]):
loss_D_llike += tf.reduce_sum(
0.5 * tf.squared_difference(utils.flatten(h_fake),
utils.flatten(h_real)), 1)
# GAN Loss
eps = 1e-12
loss_real = tf.reduce_sum(tf.log(D_real['probs'] + eps), 1)
loss_fake = tf.reduce_sum(tf.log(1 - D_fake['probs'] + eps), 1)
loss_samp = tf.reduce_sum(tf.log(1 - D_samp['probs'] + eps), 1)
loss_GAN = (loss_real + loss_fake + loss_samp) / 3.0
loss_enc = tf.reduce_mean(loss_z + loss_D_llike)
loss_gen = tf.reduce_mean(gamma * loss_D_llike - loss_GAN)
loss_dis = -tf.reduce_mean(loss_GAN)
return {
'x': x,
'z': z,
'x_tilde': decoding_actual['x_tilde'],
'z_samp': z_samp,
'x_tilde_samp': decoding_sampled['x_tilde'],
'loss_real': loss_real,
'loss_fake': loss_fake,
'loss_samp': loss_samp,
'loss_GAN': loss_GAN,
'loss_D_llike': loss_D_llike,
'loss_enc': loss_enc,
'loss_gen': loss_gen,
'loss_dis': loss_dis,
'gamma': gamma
}
def encoder(x,
n_hidden=None,
dimensions=[],
filter_sizes=[],
convolutional=False,
activation=tf.nn.relu,
output_activation=tf.nn.sigmoid):
"""Summary
Parameters
----------
x : TYPE
Description
n_hidden : None, optional
Description
dimensions : list, optional
Description
filter_sizes : list, optional
Description
convolutional : bool, optional
Description
activation : TYPE, optional
Description
output_activation : TYPE, optional
Description
Returns
-------
name : TYPE
Description
"""
if convolutional:
x_tensor = utils.to_tensor(x)
else:
x_tensor = tf.reshape(tensor=x, shape=[-1, dimensions[0]])
dimensions = dimensions[1:]
current_input = x_tensor
Ws = []
hs = []
shapes = []
for layer_i, n_output in enumerate(dimensions):
with tf.variable_scope(str(layer_i)):
shapes.append(current_input.get_shape().as_list())
if convolutional:
h, W = utils.conv2d(x=current_input,
n_output=n_output,
k_h=filter_sizes[layer_i],
k_w=filter_sizes[layer_i],
padding='SAME')
else:
h, W = utils.linear(x=current_input, n_output=n_output)
h = activation(h)
Ws.append(W)
hs.append(h)
current_input = h
shapes.append(h.get_shape().as_list())
with tf.variable_scope('flatten'):
flattened = utils.flatten(current_input)
with tf.variable_scope('hidden'):
if n_hidden:
h, W = utils.linear(flattened, n_hidden, name='linear')
h = activation(h)
else:
h = flattened
return {'z': h, 'Ws': Ws, 'hs': hs, 'shapes': shapes}
def VAE(input_shape=[None, 784],
n_filters=[64, 64, 64],
filter_sizes=[4, 4, 4],
n_hidden=32,
n_code=2,
activation=tf.nn.tanh,
convolutional=False,
variational=False):
"""Summary
Parameters
----------
input_shape : list, optional
Description
n_filters : list, optional
Description
filter_sizes : list, optional
Description
n_hidden : int, optional
Description
n_code : int, optional
Description
activation : TYPE, optional
Description
convolutional : bool, optional
Description
variational : bool, optional
Description
Returns
-------
name : TYPE
Description
"""
# network input / placeholders for train (bn)
x = tf.placeholder(tf.float32, input_shape, 'x')
with tf.variable_scope('encoder'):
encoding = encoder(x=x,
n_hidden=n_hidden,
convolutional=convolutional,
dimensions=n_filters,
filter_sizes=filter_sizes,
activation=activation)
if variational:
with tf.variable_scope('variational'):
z, z_mu, z_log_sigma, loss_z = variational_bayes(h=encoding['z'],
n_code=n_code)
else:
z = encoding['z']
loss_z = None
shapes = encoding['shapes'].copy()
shapes.reverse()
n_filters = n_filters.copy()
n_filters.reverse()
n_filters += [input_shape[-1]]
with tf.variable_scope('generator'):
decoding = decoder(z=z,
shapes=shapes,
n_hidden=n_hidden,
dimensions=n_filters,
filter_sizes=filter_sizes,
convolutional=convolutional,
activation=activation)
x_tilde = decoding['x_tilde']
x_flat = utils.flatten(x)
x_tilde_flat = utils.flatten(x_tilde)
# -log(p(x|z))
loss_x = tf.reduce_sum(tf.squared_difference(x_flat, x_tilde_flat), 1)
return {
'loss_x': loss_x,
'loss_z': loss_z,
'x': x,
'z': z,
'Ws': encoding['Ws'],
'hs': decoding['hs'],
'x_tilde': x_tilde
}
def test_flatten():
assert(utils.flatten(
tf.constant(np.zeros((3, 100, 100, 3)))).get_shape().as_list() == [3, 30000])
def fcn_24_detect(threshold, dropout=False, activation=tf.nn.relu):
imgs = tf.placeholder(tf.float32, [None, 24, 24, 3])
labels = tf.placeholder(tf.float32, [None, 1])
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
net_12 = fcn_12_detect(0.16, activation=activation)
with tf.variable_scope('net_24'):
conv1, _ = utils.conv2d(x=imgs,
n_output=64,
k_w=5,
k_h=5,
d_w=1,
d_h=1,
name="conv1")
conv1 = activation(conv1)
pool1 = tf.nn.max_pool(conv1,
ksize=[1, 3, 3, 1],
strides=[1, 2, 2, 1],
padding="SAME",
name="pool1")
ip1, W1 = utils.conv2d(x=pool1,
n_output=128,
k_w=12,
k_h=12,
d_w=1,
d_h=1,
padding="VALID",
name="ip1")
ip1 = activation(ip1)
net_12_ip1 = net_12['features']
concat = tf.concat(3, [ip1, net_12_ip1])
if dropout:
concat = tf.nn.dropout(concat, keep_prob)
ip2, W2 = utils.conv2d(x=concat,
n_output=1,
k_w=1,
k_h=1,
d_w=1,
d_h=1,
name="ip2")
pred = tf.nn.sigmoid(utils.flatten(ip2))
target = utils.flatten(labels)
regularizer = 8e-3 * (tf.nn.l2_loss(W1) + 100 * tf.nn.l2_loss(W2))
loss = tf.reduce_mean(
tf.div(
tf.add(
-tf.reduce_sum(target * tf.log(pred + 1e-9), 1),
-tf.reduce_sum((1 - target) * tf.log(1 - pred + 1e-9), 1)),
2)) + regularizer
cost = tf.reduce_mean(loss)
thresholding_24 = tf.cast(tf.greater(pred, threshold), "float")
recall_24 = tf.reduce_sum(
tf.cast(
tf.logical_and(tf.equal(thresholding_24, tf.constant([1.0])),
tf.equal(target, tf.constant([1.0]))),
"float")) / tf.reduce_sum(target)
correct_prediction = tf.equal(
tf.cast(tf.greater(pred, threshold), tf.int32),
tf.cast(target, tf.int32))
acc = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
return {
'net_12': net_12,
'imgs': imgs,
'labels': labels,
'imgs_12': net_12['imgs'],
'labels_12': net_12['labels'],
'keep_prob': keep_prob,
'keep_prob_12': net_12['keep_prob'],
'cost': cost,
'pred': pred,
'accuracy': acc,
'features': concat,
'recall': recall_24,
'thresholding': thresholding_24
}
def VAE(input_shape=[None, 784],
n_filters=[64, 64, 64],
filter_sizes=[4, 4, 4],
n_hidden=32,
n_code=2,
activation=tf.nn.tanh,
dropout=False,
denoising=False,
convolutional=False,
variational=False):
"""(Variational) (Convolutional) (Denoising) Autoencoder.
Uses tied weights.
Parameters
----------
input_shape : list, optional
Shape of the input to the network. e.g. for MNIST: [None, 784].
n_filters : list, optional
Number of filters for each layer.
If convolutional=True, this refers to the total number of output
filters to create for each layer, with each layer's number of output
filters as a list.
If convolutional=False, then this refers to the total number of neurons
for each layer in a fully connected network.
filter_sizes : list, optional
Only applied when convolutional=True. This refers to the ksize (height
and width) of each convolutional layer.
n_hidden : int, optional
Only applied when variational=True. This refers to the first fully
connected layer prior to the variational embedding, directly after
the encoding. After the variational embedding, another fully connected
layer is created with the same size prior to decoding. Set to 0 to
not use an additional hidden layer.
n_code : int, optional
Only applied when variational=True. This refers to the number of
latent Gaussians to sample for creating the inner most encoding.
activation : function, optional
Activation function to apply to each layer, e.g. tf.nn.relu
dropout : bool, optional
Whether or not to apply dropout. If using dropout, you must feed a
value for 'keep_prob', as returned in the dictionary. 1.0 means no
dropout is used. 0.0 means every connection is dropped. Sensible
values are between 0.5-0.8.
denoising : bool, optional
Whether or not to apply denoising. If using denoising, you must feed a
value for 'corrupt_prob', as returned in the dictionary. 1.0 means no
corruption is used. 0.0 means every feature is corrupted. Sensible
values are between 0.5-0.8.
convolutional : bool, optional
Whether or not to use a convolutional network or else a fully connected
network will be created. This effects the n_filters parameter's
meaning.
variational : bool, optional
Whether or not to create a variational embedding layer. This will
create a fully connected layer after the encoding, if `n_hidden` is
greater than 0, then will create a multivariate gaussian sampling
layer, then another fully connected layer. The size of the fully
connected layers are determined by `n_hidden`, and the size of the
sampling layer is determined by `n_code`.
Returns
-------
model : dict
{
'cost': Tensor to optimize.
'Ws': All weights of the encoder.
'x': Input Placeholder
'z': Inner most encoding Tensor (latent features)
'y': Reconstruction of the Decoder
'keep_prob': Amount to keep when using Dropout
'corrupt_prob': Amount to corrupt when using Denoising
'train': Set to True when training/Applies to Batch Normalization.
}
"""
# network input / placeholders for train (bn) and dropout
x = tf.placeholder(tf.float32, input_shape, 'x')
phase_train = tf.placeholder(tf.bool, name='phase_train')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
corrupt_prob = tf.placeholder(tf.float32, [1])
# apply noise if denoising
x_ = (utils.corrupt(x) * corrupt_prob + x * (1 - corrupt_prob)) if denoising else x
# 2d -> 4d if convolution
x_tensor = utils.to_tensor(x_) if convolutional else x_
current_input = x_tensor
Ws = []
shapes = []
# Build the encoder
for layer_i, n_output in enumerate(n_filters):
with tf.variable_scope('encoder/{}'.format(layer_i)):
shapes.append(current_input.get_shape().as_list())
if convolutional:
h, W = utils.conv2d(x=current_input,
n_output=n_output,
k_h=filter_sizes[layer_i],
k_w=filter_sizes[layer_i])
else:
h, W = utils.linear(x=current_input,
n_output=n_output)
h = activation(batch_norm(h, phase_train, 'bn' + str(layer_i)))
if dropout:
h = tf.nn.dropout(h, keep_prob)
Ws.append(W)
current_input = h
shapes.append(current_input.get_shape().as_list())
with tf.variable_scope('variational'):
if variational:
dims = current_input.get_shape().as_list()
flattened = utils.flatten(current_input)
if n_hidden:
h = utils.linear(flattened, n_hidden, name='W_fc')[0]
h = activation(batch_norm(h, phase_train, 'fc/bn'))
if dropout:
h = tf.nn.dropout(h, keep_prob)
else:
h = flattened
z_mu = utils.linear(h, n_code, name='mu')[0]
z_log_sigma = 0.5 * utils.linear(h, n_code, name='log_sigma')[0]
# Sample from noise distribution p(eps) ~ N(0, 1)
epsilon = tf.random_normal(
tf.stack([tf.shape(x)[0], n_code]))
# Sample from posterior
z = z_mu + tf.multiply(epsilon, tf.exp(z_log_sigma))
if n_hidden:
h = utils.linear(z, n_hidden, name='fc_t')[0]
h = activation(batch_norm(h, phase_train, 'fc_t/bn'))
if dropout:
h = tf.nn.dropout(h, keep_prob)
else:
h = z
size = dims[1] * dims[2] * dims[3] if convolutional else dims[1]
h = utils.linear(h, size, name='fc_t2')[0]
current_input = activation(batch_norm(h, phase_train, 'fc_t2/bn'))
if dropout:
current_input = tf.nn.dropout(current_input, keep_prob)
if convolutional:
current_input = tf.reshape(
current_input, tf.stack([
tf.shape(current_input)[0],
dims[1],
dims[2],
dims[3]]))
else:
z = current_input
shapes.reverse()
n_filters.reverse()
Ws.reverse()
n_filters += [input_shape[-1]]
# %%
# Decoding layers
for layer_i, n_output in enumerate(n_filters[1:]):
with tf.variable_scope('decoder/{}'.format(layer_i)):
shape = shapes[layer_i + 1]
if convolutional:
h, W = utils.deconv2d(x=current_input,
n_output_h=shape[1],
n_output_w=shape[2],
n_output_ch=shape[3],
n_input_ch=shapes[layer_i][3],
k_h=filter_sizes[layer_i],
k_w=filter_sizes[layer_i])
else:
h, W = utils.linear(x=current_input,
n_output=n_output)
h = activation(batch_norm(h, phase_train, 'dec/bn' + str(layer_i)))
if dropout:
h = tf.nn.dropout(h, keep_prob)
current_input = h
y = current_input
x_flat = utils.flatten(x)
y_flat = utils.flatten(y)
# l2 loss
loss_x = tf.reduce_sum(tf.squared_difference(x_flat, y_flat), 1)
if variational:
# variational lower bound, kl-divergence
loss_z = -0.5 * tf.reduce_sum(
1.0 + 2.0 * z_log_sigma -
tf.square(z_mu) - tf.exp(2.0 * z_log_sigma), 1)
# add l2 loss
cost = tf.reduce_mean(loss_x + loss_z)
else:
# just optimize l2 loss
cost = tf.reduce_mean(loss_x)
return {'cost': cost, 'Ws': Ws,
'x': x, 'z': z, 'y': y,
'keep_prob': keep_prob,
'corrupt_prob': corrupt_prob,
'train': phase_train}
best_descs = ['haddad_desc', 'all', 'vib_100']
nice_names = {'none_opt': 'default parameters (q2)',
'k_opt': 'k_best optimized (q2)',
'reg_opt': 'C optimized (q2)',
'both_opt': 'both optimized (q2)',
}
line_max = 0.9
ticks = [0, 0.2, 0.4, 0.6, 0.8, 0.9]
ticklabels = ['0', '.2', '.4', '.6', '.8', '']
vals = [out_res[k] for k in best_descs]
reference_name = 'none_opt'
to_compare = ['k_opt', 'reg_opt'] #, 'both_opt']
titles = ['c)', 'd)']
for i, pick_type in enumerate(to_compare):
reference = np.array(utils.flatten([r[reference_name] for r in vals]))
improved = np.array(utils.flatten([r[pick_type] for r in vals]))
# only plot values which reach genscore > 0 after paramsearch
reference = reference[improved > 0]
improved = improved[improved > 0]
# don't care how negative it was before search, all < 0 are equally bad
reference[reference < 0] = 0
improved[improved < 0] = 0
# fig = plt.figure(figsize=(2.23, 2))
ax = plt.subplot(gs[0,i])#fig.add_subplot(1, 3, i+1)
ax.plot(reference, improved, 'ko', alpha=0.6, markersize=4)
ax.plot([0, line_max-0.05], [0, line_max-0.05], color='0.5')
ax.set_ylabel(nice_names[pick_type])
plt.axis('scaled')
ax.set_xlim([-0.05, line_max])
def deepID(input_shape=[None, 39, 39, 1],
n_filters=[20, 40, 60, 80],
filter_sizes=[4, 3, 3, 2],
activation=tf.nn.relu,
dropout=False):
"""DeepID.
Uses tied weights.
Parameters
----------
input_shape : list, optional
Shape of the input to the network. e.g. for MNIST: [None, 784].
n_filters : list, optional
Number of filters for each layer.
If convolutional=True, this refers to the total number of output
filters to create for each layer, with each layer's number of output
filters as a list.
If convolutional=False, then this refers to the total number of neurons
for each layer in a fully connected network.
filter_sizes : list, optional
Only applied when convolutional=True. This refers to the ksize (height
and width) of each convolutional layer.
activation : function, optional
Activation function to apply to each layer, e.g. tf.nn.relu
dropout : bool, optional
Whether or not to apply dropout. If using dropout, you must feed a
value for 'keep_prob', as returned in the dictionary. 1.0 means no
dropout is used. 0.0 means every connection is dropped. Sensible
values are between 0.5-0.8.
Returns
-------
model : dict
{
'cost': Tensor to optimize.
'Ws': All weights of the encoder.
'x': Input Placeholder
'z': Inner most encoding Tensor (latent features)
'y': Reconstruction of the Decoder
'keep_prob': Amount to keep when using Dropout
'corrupt_prob': Amount to corrupt when using Denoising
'train': Set to True when training/Applies to Batch Normalization.
}
"""
# network input / placeholders for train (bn) and dropout
x = tf.placeholder(tf.float32, input_shape, 'x')
y = tf.placeholder(tf.float32, [None, 10], 'y')
phase_train = tf.placeholder(tf.bool, name='phase_train')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
# 2d -> 4d if convolution
x_tensor = utils.to_tensor(x)
current_input = x_tensor
Ws = []
shapes = []
# Build the encoder
shapes.append(current_input.get_shape().as_list())
conv1, W = utils.conv2d(x=x_tensor,
n_output=n_filters[0],
k_h=filter_sizes[0],
k_w=filter_sizes[0],
d_w=1,
d_h=1,
name='conv1')
Ws.append(W)
# conv1 = activation(batch_norm(conv1, phase_train, 'bn1'))
conv1 = activation(conv1)
pool1 = tf.nn.max_pool(conv1,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool1')
conv2, W = utils.conv2d(x=pool1,
n_output=n_filters[1],
k_h=filter_sizes[1],
k_w=filter_sizes[1],
d_w=1,
d_h=1,
name='conv2')
Ws.append(W)
# conv2 = activation(batch_norm(conv2, phase_train, 'bn2'))
conv2 = activation(conv2)
pool2 = tf.nn.max_pool(conv2,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool2')
conv3, W = utils.conv2d(x=pool2,
n_output=n_filters[2],
k_h=filter_sizes[2],
k_w=filter_sizes[2],
d_w=1,
d_h=1,
name='conv3')
Ws.append(W)
# conv3 = activation(batch_norm(conv3, phase_train, 'bn3'))
conv3 = activation(conv3)
pool3 = tf.nn.max_pool(conv3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool3')
conv4, W = utils.conv2d(x=pool3,
n_output=n_filters[3],
k_h=filter_sizes[3],
k_w=filter_sizes[3],
d_w=1,
d_h=1,
name='conv4')
Ws.append(W)
# conv4 = activation(batch_norm(conv4, phase_train, 'bn4'))
conv4 = activation(conv4)
pool3_flat = utils.flatten(pool3)
conv4_flat = utils.flatten(conv4)
concat = tf.concat(1, [pool3_flat, conv4_flat], name='concat')
ip1, W = utils.linear(concat, 120, name='ip1')
Ws.append(W)
ip1 = activation(ip1)
if dropout:
ip1 = tf.nn.dropout(ip1, keep_prob)
ip2, W = utils.linear(ip1, 10, name='ip2')
Ws.append(W)
# ip2 = activation(ip2)
p_flat = utils.flatten(ip2)
y_flat = utils.flatten(y)
regularizers = 5e-4 * (tf.nn.l2_loss(Ws[-1]) + tf.nn.l2_loss(Ws[-2]))
# l2 loss
loss_x = tf.reduce_sum(tf.squared_difference(p_flat, y_flat), 1)
cost = tf.reduce_mean(loss_x) + regularizers
prediction = tf.reshape(p_flat, (-1, 5, 2))
return {
'cost': cost,
'Ws': Ws,
'x': x,
'y': y,
'pred': prediction,
'keep_prob': keep_prob,
'train': phase_train
}
def VAEGAN(input_shape=[None, 784],
n_filters=[64, 64, 64],
filter_sizes=[4, 4, 4],
n_hidden=32,
n_code=2,
activation=tf.nn.tanh,
convolutional=False,
variational=False):
"""Summary
Parameters
----------
input_shape : list, optional
Description
n_filters : list, optional
Description
filter_sizes : list, optional
Description
n_hidden : int, optional
Description
n_code : int, optional
Description
activation : TYPE, optional
Description
convolutional : bool, optional
Description
variational : bool, optional
Description
Returns
-------
name : TYPE
Description
"""
# network input / placeholders for train (bn)
x = tf.placeholder(tf.float32, input_shape, 'x')
z_samp = tf.placeholder(tf.float32, [None, n_code], 'z_samp')
with tf.variable_scope('encoder'):
encoding = encoder(
x=x,
n_hidden=n_hidden,
convolutional=convolutional,
dimensions=n_filters,
filter_sizes=filter_sizes,
activation=activation)
with tf.variable_scope('variational'):
z, z_mu, z_log_sigma, loss_z = variational_bayes(
h=encoding['z'], n_code=n_code)
shapes = encoding['shapes'].copy()
shapes.reverse()
n_filters_decoder = n_filters.copy()
n_filters_decoder.reverse()
n_filters_decoder += [input_shape[-1]]
with tf.variable_scope('generator'):
decoding_actual = decoder(
z=z,
shapes=shapes,
n_hidden=n_hidden,
convolutional=convolutional,
dimensions=n_filters_decoder,
filter_sizes=filter_sizes,
activation=activation)
with tf.variable_scope('generator', reuse=True):
decoding_sampled = decoder(
z=z_samp,
shapes=shapes,
n_hidden=n_hidden,
convolutional=convolutional,
dimensions=n_filters_decoder,
filter_sizes=filter_sizes,
activation=activation)
with tf.variable_scope('discriminator'):
D_real = discriminator(
x,
filter_sizes=filter_sizes,
n_filters=n_filters,
activation=activation)
with tf.variable_scope('discriminator', reuse=True):
D_fake = discriminator(
decoding_actual['x_tilde'],
filter_sizes=filter_sizes,
n_filters=n_filters,
activation=activation)
with tf.variable_scope('discriminator', reuse=True):
D_samp = discriminator(
decoding_sampled['x_tilde'],
filter_sizes=filter_sizes,
n_filters=n_filters,
activation=activation)
with tf.variable_scope('loss'):
# Weights influence of content/style of decoder
gamma = tf.placeholder(tf.float32, name='gamma')
# Discriminator_l Log Likelihood Loss
loss_D_llike = 0
for h_fake, h_real in zip(D_fake['hs'][3:], D_real['hs'][3:]):
loss_D_llike += tf.reduce_sum(0.5 * tf.squared_difference(
utils.flatten(h_fake), utils.flatten(h_real)), 1)
# GAN Loss
eps = 1e-12
loss_real = tf.reduce_sum(tf.log(D_real['probs'] + eps), 1)
loss_fake = tf.reduce_sum(tf.log(1 - D_fake['probs'] + eps), 1)
loss_samp = tf.reduce_sum(tf.log(1 - D_samp['probs'] + eps), 1)
loss_GAN = (loss_real + loss_fake + loss_samp) / 3.0
loss_enc = tf.reduce_mean(loss_z + loss_D_llike)
loss_gen = tf.reduce_mean(gamma * loss_D_llike - loss_GAN)
loss_dis = -tf.reduce_mean(loss_GAN)
return {
'x': x,
'z': z,
'x_tilde': decoding_actual['x_tilde'],
'z_samp': z_samp,
'x_tilde_samp': decoding_sampled['x_tilde'],
'loss_real': loss_real,
'loss_fake': loss_fake,
'loss_samp': loss_samp,
'loss_GAN': loss_GAN,
'loss_D_llike': loss_D_llike,
'loss_enc': loss_enc,
'loss_gen': loss_gen,
'loss_dis': loss_dis,
'gamma': gamma
}
def encoder(x,
n_hidden=None,
dimensions=[],
filter_sizes=[],
convolutional=False,
activation=tf.nn.relu,
output_activation=tf.nn.sigmoid):
"""Summary
Parameters
----------
x : TYPE
Description
n_hidden : None, optional
Description
dimensions : list, optional
Description
filter_sizes : list, optional
Description
convolutional : bool, optional
Description
activation : TYPE, optional
Description
output_activation : TYPE, optional
Description
Returns
-------
name : TYPE
Description
"""
if convolutional:
x_tensor = utils.to_tensor(x)
else:
x_tensor = tf.reshape(tensor=x, shape=[-1, dimensions[0]])
dimensions = dimensions[1:]
current_input = x_tensor
Ws = []
hs = []
shapes = []
for layer_i, n_output in enumerate(dimensions):
with tf.variable_scope(str(layer_i)):
shapes.append(current_input.get_shape().as_list())
if convolutional:
h, W = utils.conv2d(
x=current_input,
n_output=n_output,
k_h=filter_sizes[layer_i],
k_w=filter_sizes[layer_i],
padding='SAME')
else:
h, W = utils.linear(x=current_input, n_output=n_output)
h = activation(h)
Ws.append(W)
hs.append(h)
current_input = h
shapes.append(h.get_shape().as_list())
with tf.variable_scope('flatten'):
flattened = utils.flatten(current_input)
with tf.variable_scope('hidden'):
if n_hidden:
h, W = utils.linear(flattened, n_hidden, name='linear')
h = activation(h)
else:
h = flattened
return {'z': h, 'Ws': Ws, 'hs': hs, 'shapes': shapes}
def VAE(input_shape=[None, 784],
n_filters=[64, 64, 64],
filter_sizes=[4, 4, 4],
n_hidden=32,
n_code=2,
activation=tf.nn.tanh,
convolutional=False,
variational=False):
"""Summary
Parameters
----------
input_shape : list, optional
Description
n_filters : list, optional
Description
filter_sizes : list, optional
Description
n_hidden : int, optional
Description
n_code : int, optional
Description
activation : TYPE, optional
Description
convolutional : bool, optional
Description
variational : bool, optional
Description
Returns
-------
name : TYPE
Description
"""
# network input / placeholders for train (bn)
x = tf.placeholder(tf.float32, input_shape, 'x')
with tf.variable_scope('encoder'):
encoding = encoder(
x=x,
n_hidden=n_hidden,
convolutional=convolutional,
dimensions=n_filters,
filter_sizes=filter_sizes,
activation=activation)
if variational:
with tf.variable_scope('variational'):
z, z_mu, z_log_sigma, loss_z = variational_bayes(
h=encoding['z'], n_code=n_code)
else:
z = encoding['z']
loss_z = None
shapes = encoding['shapes'].copy()
shapes.reverse()
n_filters = n_filters.copy()
n_filters.reverse()
n_filters += [input_shape[-1]]
with tf.variable_scope('generator'):
decoding = decoder(
z=z,
shapes=shapes,
n_hidden=n_hidden,
dimensions=n_filters,
filter_sizes=filter_sizes,
convolutional=convolutional,
activation=activation)
x_tilde = decoding['x_tilde']
x_flat = utils.flatten(x)
x_tilde_flat = utils.flatten(x_tilde)
# -log(p(x|z))
loss_x = tf.reduce_sum(tf.squared_difference(x_flat, x_tilde_flat), 1)
return {
'loss_x': loss_x,
'loss_z': loss_z,
'x': x,
'z': z,
'Ws': encoding['Ws'],
'hs': decoding['hs'],
'x_tilde': x_tilde
}
def VAE(input_shape=[None, 784],
n_filters=[64, 64, 64],
filter_sizes=[4, 4, 4],
n_hidden=32,
n_code=2,
activation=tf.nn.tanh,
dropout=False,
denoising=False,
convolutional=False,
variational=False):
"""(Variational) (Convolutional) (Denoising) Autoencoder.
Uses tied weights.
Parameters
----------
input_shape : list, optional
Shape of the input to the network. e.g. for MNIST: [None, 784].
n_filters : list, optional
Number of filters for each layer.
If convolutional=True, this refers to the total number of output
filters to create for each layer, with each layer's number of output
filters as a list.
If convolutional=False, then this refers to the total number of neurons
for each layer in a fully connected network.
filter_sizes : list, optional
Only applied when convolutional=True. This refers to the ksize (height
and width) of each convolutional layer.
n_hidden : int, optional
Only applied when variational=True. This refers to the first fully
connected layer prior to the variational embedding, directly after
the encoding. After the variational embedding, another fully connected
layer is created with the same size prior to decoding. Set to 0 to
not use an additional hidden layer.
n_code : int, optional
Only applied when variational=True. This refers to the number of
latent Gaussians to sample for creating the inner most encoding.
activation : function, optional
Activation function to apply to each layer, e.g. tf.nn.relu
dropout : bool, optional
Whether or not to apply dropout. If using dropout, you must feed a
value for 'keep_prob', as returned in the dictionary. 1.0 means no
dropout is used. 0.0 means every connection is dropped. Sensible
values are between 0.5-0.8.
denoising : bool, optional
Whether or not to apply denoising. If using denoising, you must feed a
value for 'corrupt_prob', as returned in the dictionary. 1.0 means no
corruption is used. 0.0 means every feature is corrupted. Sensible
values are between 0.5-0.8.
convolutional : bool, optional
Whether or not to use a convolutional network or else a fully connected
network will be created. This effects the n_filters parameter's
meaning.
variational : bool, optional
Whether or not to create a variational embedding layer. This will
create a fully connected layer after the encoding, if `n_hidden` is
greater than 0, then will create a multivariate gaussian sampling
layer, then another fully connected layer. The size of the fully
connected layers are determined by `n_hidden`, and the size of the
sampling layer is determined by `n_code`.
Returns
-------
model : dict
{
'cost': Tensor to optimize.
'Ws': All weights of the encoder.
'x': Input Placeholder
'z': Inner most encoding Tensor (latent features)
'y': Reconstruction of the Decoder
'keep_prob': Amount to keep when using Dropout
'corrupt_prob': Amount to corrupt when using Denoising
'train': Set to True when training/Applies to Batch Normalization.
}
"""
# network input / placeholders for train (bn) and dropout
x = tf.placeholder(tf.float32, input_shape, 'x')
phase_train = tf.placeholder(tf.bool, name='phase_train')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
corrupt_prob = tf.placeholder(tf.float32, [1])
# apply noise if denoising
x_ = (utils.corrupt(x) * corrupt_prob + x *
(1 - corrupt_prob)) if denoising else x
# 2d -> 4d if convolution
x_tensor = utils.to_tensor(x_) if convolutional else x_
current_input = x_tensor
Ws = []
shapes = []
# Build the encoder
for layer_i, n_output in enumerate(n_filters):
with tf.variable_scope('encoder/{}'.format(layer_i)):
shapes.append(current_input.get_shape().as_list())
if convolutional:
h, W = utils.conv2d(x=current_input,
n_output=n_output,
k_h=filter_sizes[layer_i],
k_w=filter_sizes[layer_i])
else:
h, W = utils.linear(x=current_input, n_output=n_output)
h = activation(batch_norm(h, phase_train, 'bn' + str(layer_i)))
if dropout:
h = tf.nn.dropout(h, keep_prob)
Ws.append(W)
current_input = h
shapes.append(current_input.get_shape().as_list())
with tf.variable_scope('variational'):
if variational:
dims = current_input.get_shape().as_list()
flattened = utils.flatten(current_input)
if n_hidden:
h = utils.linear(flattened, n_hidden, name='W_fc')[0]
h = activation(batch_norm(h, phase_train, 'fc/bn'))
if dropout:
h = tf.nn.dropout(h, keep_prob)
else:
h = flattened
z_mu = utils.linear(h, n_code, name='mu')[0]
z_log_sigma = 0.5 * utils.linear(h, n_code, name='log_sigma')[0]
# Sample from noise distribution p(eps) ~ N(0, 1)
epsilon = tf.random_normal(tf.stack([tf.shape(x)[0], n_code]))
# Sample from posterior
z = z_mu + tf.multiply(epsilon, tf.exp(z_log_sigma))
if n_hidden:
h = utils.linear(z, n_hidden, name='fc_t')[0]
h = activation(batch_norm(h, phase_train, 'fc_t/bn'))
if dropout:
h = tf.nn.dropout(h, keep_prob)
else:
h = z
size = dims[1] * dims[2] * dims[3] if convolutional else dims[1]
h = utils.linear(h, size, name='fc_t2')[0]
current_input = activation(batch_norm(h, phase_train, 'fc_t2/bn'))
if dropout:
current_input = tf.nn.dropout(current_input, keep_prob)
if convolutional:
current_input = tf.reshape(
current_input,
tf.stack([
tf.shape(current_input)[0], dims[1], dims[2], dims[3]
]))
else:
z = current_input
shapes.reverse()
n_filters.reverse()
Ws.reverse()
n_filters += [input_shape[-1]]
# %%
# Decoding layers
for layer_i, n_output in enumerate(n_filters[1:]):
with tf.variable_scope('decoder/{}'.format(layer_i)):
shape = shapes[layer_i + 1]
if convolutional:
h, W = utils.deconv2d(x=current_input,
n_output_h=shape[1],
n_output_w=shape[2],
n_output_ch=shape[3],
n_input_ch=shapes[layer_i][3],
k_h=filter_sizes[layer_i],
k_w=filter_sizes[layer_i])
else:
h, W = utils.linear(x=current_input, n_output=n_output)
h = activation(batch_norm(h, phase_train, 'dec/bn' + str(layer_i)))
if dropout:
h = tf.nn.dropout(h, keep_prob)
current_input = h
y = current_input
x_flat = utils.flatten(x)
y_flat = utils.flatten(y)
# l2 loss
loss_x = tf.reduce_sum(tf.squared_difference(x_flat, y_flat), 1)
if variational:
# variational lower bound, kl-divergence
loss_z = -0.5 * tf.reduce_sum(
1.0 + 2.0 * z_log_sigma - tf.square(z_mu) -
tf.exp(2.0 * z_log_sigma), 1)
# add l2 loss
cost = tf.reduce_mean(loss_x + loss_z)
else:
# just optimize l2 loss
cost = tf.reduce_mean(loss_x)
return {
'cost': cost,
'Ws': Ws,
'x': x,
'z': z,
'y': y,
'keep_prob': keep_prob,
'corrupt_prob': corrupt_prob,
'train': phase_train
}
def fcn_48_cal(dropout=False, activation=tf.nn.relu):
imgs = tf.placeholder(tf.float32, [None, 48, 48, 3])
labels = tf.placeholder(tf.float32, [None])
with tf.variable_scope('cal_48'):
conv1, _ = utils.conv2d(x=imgs,
n_output=64,
k_w=5,
k_h=5,
d_w=1,
d_h=1,
name="conv1")
conv1 = activation(conv1)
pool1 = tf.nn.max_pool(conv1,
ksize=[1, 3, 3, 1],
strides=[1, 2, 2, 1],
padding="SAME",
name="pool1")
conv2, _ = utils.conv2d(x=pool1,
n_output=64,
k_w=5,
k_h=5,
d_w=1,
d_h=1,
name="conv2")
ip1, W1 = utils.conv2d(x=conv2,
n_output=256,
k_w=24,
k_h=24,
d_w=1,
d_h=1,
padding="VALID",
name="ip1")
ip1 = activation(ip1)
if dropout:
ip1 = tf.nn.dropout(ip1, keep_prob)
ip2, W2 = utils.conv2d(x=ip1,
n_output=45,
k_w=1,
k_h=1,
d_w=1,
d_h=1,
name="ip2")
pred = utils.flatten(ip2)
# target = utils.flatten(labels)
# label_shape = labels.get_shape().as_list()
# target = tf.reshape(labels,[label_shape[0]])
target = labels
cross_entropy = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(
pred, tf.cast(target, tf.int64)))
regularizer = 8e-3 * (tf.nn.l2_loss(W1) + 100 * tf.nn.l2_loss(W2))
loss = cross_entropy + regularizer
correct_prediction = tf.equal(tf.argmax(pred, 1),
tf.cast(target, tf.int64))
acc = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
return {
'cost': loss,
'pred': pred,
'accuracy': acc,
'target': target,
'imgs': imgs,
'labels': labels
}
def VAE(input_shape=[None, 784],
output_shape=[None, 784],
n_filters=[64, 64, 64],
filter_sizes=[4, 4, 4],
n_hidden=32,
n_code=2,
activation=tf.nn.tanh,
dropout=False,
denoising=False,
convolutional=False,
variational=False,
softmax=False,
classifier='alexnet_v2'):
"""(Variational) (Convolutional) (Denoising) Autoencoder.
Uses tied weights.
Parameters
----------
input_shape : list, optional
Shape of the input to the network. e.g. for MNIST: [None, 784].
n_filters : list, optional
Number of filters for each layer.
If convolutional=True, this refers to the total number of output
filters to create for each layer, with each layer's number of output
filters as a list.
If convolutional=False, then this refers to the total number of neurons
for each layer in a fully connected network.
filter_sizes : list, optional
Only applied when convolutional=True. This refers to the ksize (height
and width) of each convolutional layer.
n_hidden : int, optional
Only applied when variational=True. This refers to the first fully
connected layer prior to the variational embedding, directly after
the encoding. After the variational embedding, another fully connected
layer is created with the same size prior to decoding. Set to 0 to
not use an additional hidden layer.
n_code : int, optional
Only applied when variational=True. This refers to the number of
latent Gaussians to sample for creating the inner most encoding.
activation : function, optional
Activation function to apply to each layer, e.g. tf.nn.relu
dropout : bool, optional
Whether or not to apply dropout. If using dropout, you must feed a
value for 'keep_prob', as returned in the dictionary. 1.0 means no
dropout is used. 0.0 means every connection is dropped. Sensible
values are between 0.5-0.8.
denoising : bool, optional
Whether or not to apply denoising. If using denoising, you must feed a
value for 'corrupt_rec', as returned in the dictionary. 1.0 means no
corruption is used. 0.0 means every feature is corrupted. Sensible
values are between 0.5-0.8.
convolutional : bool, optional
Whether or not to use a convolutional network or else a fully connected
network will be created. This effects the n_filters parameter's
meaning.
variational : bool, optional
Whether or not to create a variational embedding layer. This will
create a fully connected layer after the encoding, if `n_hidden` is
greater than 0, then will create a multivariate gaussian sampling
layer, then another fully connected layer. The size of the fully
connected layers are determined by `n_hidden`, and the size of the
sampling layer is determined by `n_code`.
Returns
-------
model : dict
{
'cost': Tensor to optimize.
'Ws': All weights of the encoder.
'x': Input Placeholder
'z': Inner most encoding Tensor (latent features)
'y': Reconstruction of the Decoder
'keep_prob': Amount to keep when using Dropout
'corrupt_rec': Amount to corrupt when using Denoising
'train': Set to True when training/Applies to Batch Normalization.
}
"""
# network input / placeholders for train (bn) and dropout
x = tf.placeholder(tf.float32, input_shape, 'x')
t = tf.placeholder(tf.float32, output_shape, 't')
label = tf.placeholder(tf.int32, [None], 'label')
phase_train = tf.placeholder(tf.bool, name='phase_train')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
corrupt_rec = tf.placeholder(tf.float32, name='corrupt_rec')
corrupt_cls = tf.placeholder(tf.float32, name='corrupt_cls')
# input of the reconstruction network
# np.tanh(2) = 0.964
current_input1 = utils.corrupt(x)*corrupt_rec + x*(1-corrupt_rec) \
if (denoising and phase_train is not None) else x
current_input1.set_shape(x.get_shape())
# 2d -> 4d if convolution
current_input1 = utils.to_tensor(current_input1) \
if convolutional else current_input1
Ws = []
shapes = []
# Build the encoder
for layer_i, n_output in enumerate(n_filters):
with tf.variable_scope('encoder/{}'.format(layer_i)):
shapes.append(current_input1.get_shape().as_list())
if convolutional:
h, W = utils.conv2d(x=current_input1,
n_output=n_output,
k_h=filter_sizes[layer_i],
k_w=filter_sizes[layer_i])
else:
h, W = utils.linear(x=current_input1, n_output=n_output)
h = activation(batch_norm(h, phase_train, 'bn' + str(layer_i)))
if dropout:
h = tf.nn.dropout(h, keep_prob)
Ws.append(W)
current_input1 = h
shapes.append(current_input1.get_shape().as_list())
with tf.variable_scope('variational'):
if variational:
dims = current_input1.get_shape().as_list()
flattened = utils.flatten(current_input1)
if n_hidden:
h = utils.linear(flattened, n_hidden, name='W_fc')[0]
h = activation(batch_norm(h, phase_train, 'fc/bn'))
if dropout:
h = tf.nn.dropout(h, keep_prob)
else:
h = flattened
z_mu = utils.linear(h, n_code, name='mu')[0]
z_log_sigma = 0.5 * utils.linear(h, n_code, name='log_sigma')[0]
# modified by yidawang
# s, u, v = tf.svd(z_log_sigma)
# z_log_sigma = tf.matmul(
# tf.matmul(u, tf.diag(s)), tf.transpose(v))
# end yidawang
# Sample from noise distribution p(eps) ~ N(0, 1)
epsilon = tf.random_normal(tf.stack([tf.shape(x)[0], n_code]))
# Sample from posterior
z = z_mu + tf.multiply(epsilon, tf.exp(z_log_sigma))
if n_hidden:
h = utils.linear(z, n_hidden, name='fc_t')[0]
h = activation(batch_norm(h, phase_train, 'fc_t/bn'))
if dropout:
h = tf.nn.dropout(h, keep_prob)
else:
h = z
size = dims[1] * dims[2] * dims[3] if convolutional else dims[1]
h = utils.linear(h, size, name='fc_t2')[0]
current_input1 = activation(batch_norm(h, phase_train, 'fc_t2/bn'))
if dropout:
current_input1 = tf.nn.dropout(current_input1, keep_prob)
if convolutional:
current_input1 = tf.reshape(
current_input1,
tf.stack([
tf.shape(current_input1)[0], dims[1], dims[2], dims[3]
]))
else:
z = current_input1
shapes.reverse()
n_filters.reverse()
Ws.reverse()
n_filters += [input_shape[-1]]
# %%
# Decoding layers
for layer_i, n_output in enumerate(n_filters[1:]):
with tf.variable_scope('decoder/{}'.format(layer_i)):
shape = shapes[layer_i + 1]
if convolutional:
h, W = utils.deconv2d(x=current_input1,
n_output_h=shape[1],
n_output_w=shape[2],
n_output_ch=shape[3],
n_input_ch=shapes[layer_i][3],
k_h=filter_sizes[layer_i],
k_w=filter_sizes[layer_i])
else:
h, W = utils.linear(x=current_input1, n_output=n_output)
h = activation(batch_norm(h, phase_train, 'dec/bn' + str(layer_i)))
if dropout:
h = tf.nn.dropout(h, keep_prob)
current_input1 = h
y = current_input1
t_flat = utils.flatten(t)
y_flat = utils.flatten(y)
# l2 loss
loss_x = tf.reduce_mean(
tf.reduce_sum(tf.squared_difference(t_flat, y_flat), 1))
loss_z = 0
if variational:
# Variational lower bound, kl-divergence
loss_z = tf.reduce_mean(-0.5 * tf.reduce_sum(
1.0 + 2.0 * z_log_sigma - tf.square(z_mu) -
tf.exp(2.0 * z_log_sigma), 1))
# Add l2 loss
cost_vae = tf.reduce_mean(loss_x + loss_z)
else:
# Just optimize l2 loss
cost_vae = tf.reduce_mean(loss_x)
# Alexnet for clasification based on softmax using TensorFlow slim
if softmax:
axis = list(range(len(x.get_shape())))
mean1, variance1 = tf.nn.moments(t, axis) \
if (phase_train is True) else tf.nn.moments(x, axis)
mean2, variance2 = tf.nn.moments(y, axis)
var_prob = variance2 / variance1
# Input of the classification network
current_input2 = utils.corrupt(x)*corrupt_cls + \
x*(1-corrupt_cls) \
if (denoising and phase_train is True) else x
current_input2.set_shape(x.get_shape())
current_input2 = utils.to_tensor(current_input2) \
if convolutional else current_input2
y_concat = tf.concat([current_input2, y], 3)
with tf.variable_scope('deconv/concat'):
shape = shapes[layer_i + 1]
if convolutional:
# Here we set the input of classification network is
# the twice of
# the input of the reconstruction network
# 112->224 for alexNet and 150->300 for inception v3 and v4
y_concat, W = utils.deconv2d(
x=y_concat,
n_output_h=y_concat.get_shape()[1] * 2,
n_output_w=y_concat.get_shape()[1] * 2,
n_output_ch=y_concat.get_shape()[3],
n_input_ch=y_concat.get_shape()[3],
k_h=3,
k_w=3)
Ws.append(W)
# The following are optional networks for classification network
if classifier == 'squeezenet':
predictions, net = squeezenet.squeezenet(y_concat, num_classes=13)
elif classifier == 'zigzagnet':
predictions, net = squeezenet.zigzagnet(y_concat, num_classes=13)
elif classifier == 'alexnet_v2':
predictions, end_points = alexnet.alexnet_v2(y_concat,
num_classes=13)
elif classifier == 'inception_v1':
predictions, end_points = inception.inception_v1(y_concat,
num_classes=13)
elif classifier == 'inception_v2':
predictions, end_points = inception.inception_v2(y_concat,
num_classes=13)
elif classifier == 'inception_v3':
predictions, end_points = inception.inception_v3(y_concat,
num_classes=13)
label_onehot = tf.one_hot(label, 13, axis=-1, dtype=tf.int32)
cost_s = tf.losses.softmax_cross_entropy(label_onehot, predictions)
cost_s = tf.reduce_mean(cost_s)
acc = tf.nn.in_top_k(predictions, label, 1)
else:
predictions = tf.one_hot(label, 13, 1, 0)
label_onehot = tf.one_hot(label, 13, 1, 0)
cost_s = 0
acc = 0
# Using Summaries for Tensorboard
tf.summary.scalar('cost_vae', cost_vae)
tf.summary.scalar('cost_s', cost_s)
tf.summary.scalar('loss_x', loss_x)
tf.summary.scalar('loss_z', loss_z)
tf.summary.scalar('corrupt_rec', corrupt_rec)
tf.summary.scalar('corrupt_cls', corrupt_cls)
tf.summary.scalar('var_prob', var_prob)
merged = tf.summary.merge_all()
return {
'cost_vae': cost_vae,
'cost_s': cost_s,
'loss_x': loss_x,
'loss_z': loss_z,
'Ws': Ws,
'x': x,
't': t,
'label': label,
'label_onehot': label_onehot,
'predictions': predictions,
'z': z,
'y': y,
'acc': acc,
'keep_prob': keep_prob,
'corrupt_rec': corrupt_rec,
'corrupt_cls': corrupt_cls,
'var_prob': var_prob,
'train': phase_train,
'merged': merged
}
fig = plt.figure()
plot_res = {desc:{'ps': [], 'corrs': []} for desc in to_compare}
for desc in to_compare:
for glom in res[reference]:
corr, p = stats.pearsonr(res[reference][glom]['predictions'],
res[desc][glom]['predictions'])
plot_res[desc]['corrs'].append(corr)
plot_res[desc]['ps'].append(p)
for desc, pres in plot_res.items():
print '{} mean r: {:.2f}'.format(desc, np.mean(pres['corrs']))
ax = fig.add_subplot(1,2,1)
ax.hist(utils.flatten([v['ps'] for v in plot_res.values()]))
fig = plt.figure(figsize=(3.35, 1.8))
marker = ['o', 's', 'd']
xticks = [0, 0.2, 0.4, 0.6, 0.8]
xticklabels = ['0', '.2', '.4', '.6', '.8']
ax = fig.add_subplot(1,2,1)
ax.plot([-0.05, 0.8], [-0.05, 0.8], color='0.6')
col_ref = []
col_comp = []
for i, (desc, pres) in enumerate(plot_res.items()):
compare_scores = np.array([res[desc][g]['score'] for g in res[desc]])
ref_scores = np.array([res[reference][g]['score'] for g in res[reference]])
compare_scores[compare_scores < 0] = 0
ref_scores[ref_scores < 0] = 0
col_ref.extend(ref_scores)
