def mnist_data_providers(batch_size, crop_size=[], use_test_set=False):
if isinstance(crop_size, int): crop_size = [crop_size]
from load import mnist_with_valid_set
trX, vaX, teX, trY, vaY, teY = mnist_with_valid_set()
if use_test_set:
trX = np.concatenate([trX, vaX], axis=0)
trY = np.concatenate([trY, vaY], axis=0)
vaX = teX
vaY = teY
shape = 1, 28, 28
return {
'train':
MemoryDataProvider(trX,
trY,
batch_size,
crop_size=crop_size,
image_shape=shape),
'val':
MemoryDataProvider(vaX,
vaY,
batch_size,
crop_size=crop_size,
image_shape=shape),
'test':
MemoryDataProvider(teX,
teY,
batch_size,
crop_size=crop_size,
image_shape=shape),
}
learning_rate = 0.0002
batch_size = 128
image_shape = [28, 28, 1]
dim_z = 100
dim_W1 = 1024
dim_W2 = 128
dim_W3 = 64
dim_channel = 1
visualize_dim = 196
# SMITH: Restore model option
restore = True
end_point = 20000
trX, vaX, teX, trY, vaY, teY = mnist_with_valid_set()
dcgan_model = DCGAN(
batch_size=batch_size,
image_shape=image_shape,
dim_z=dim_z,
dim_W1=dim_W1,
dim_W2=dim_W2,
dim_W3=dim_W3,
)
Z_tf, Y_tf, image_tf, d_cost_tf, g_cost_tf, p_real, p_gen = dcgan_model.build_model(
)
sess = tf.InteractiveSession()
saver = tf.train.Saver(max_to_keep=10)
from lib.rng import py_rng, np_rng
from lib.ops import batchnorm, conv_cond_concat, deconv, dropout
from lib.theano_utils import floatX, sharedX
from lib.data_utils import shuffle, iter_data
from load import mnist_with_valid_set
#
# Phil's business
#
from MatryoshkaModules import DiscConvModule, DiscFCModule, GenConvModule, \
GenFCModule, BasicConvModule
# path for dumping experiment info and fetching dataset
EXP_DIR = "./mnist"
trX, vaX, teX, trY, vaY, teY = mnist_with_valid_set("{}/data".format(EXP_DIR))
vaX = floatX(vaX) / 255.
k = 1 # # of discrim updates for each gen update
l2 = 1.5e-5 # l2 weight decay
b1 = 0.5 # momentum term of adam
nc = 1 # # of channels in image
nbatch = 128 # # of examples in batch
npx = 28 # # of pixels width/height of images
nz0 = 32 # # of dim for Z0
nz1 = 8 # # of dim for Z1
ngfc = 256 # # of gen units for fully connected layers
ndfc = 256 # # of discrim units for fully connected layers
ngf = 64 # # of gen filters in first conv layer
ndf = 64 # # of discrim filters in first conv layer
from lib.rng import py_rng, np_rng
from lib.ops import batchnorm, conv_cond_concat, deconv, dropout
from lib.theano_utils import floatX, sharedX
from lib.data_utils import shuffle, iter_data
from load import mnist_with_valid_set
#
# Phil's business
#
from MatryoshkaModules import DiscConvModule, DiscFCModule, GenConvModule, \
GenFCModule, BasicConvModule
# path for dumping experiment info and fetching dataset
EXP_DIR = "./mnist"
trX, vaX, teX, trY, vaY, teY = mnist_with_valid_set("{}/data".format(EXP_DIR))
vaX = floatX(vaX)/255.
k = 1 # # of discrim updates for each gen update
l2 = 1.5e-5 # l2 weight decay
b1 = 0.5 # momentum term of adam
nc = 1 # # of channels in image
nbatch = 128 # # of examples in batch
npx = 28 # # of pixels width/height of images
nz0 = 32 # # of dim for Z0
nz1 = 8 # # of dim for Z1
ngfc = 256 # # of gen units for fully connected layers
ndfc = 256 # # of discrim units for fully connected layers
ngf = 64 # # of gen filters in first conv layer
ndf = 64 # # of discrim filters in first conv layer
from util import *
from load import mnist_with_valid_set
n_epochs = 10000
learning_rate = 0.0001
batch_size = 128
image_shape = [28, 28, 1]
dim_z = 100
dim_W1 = 1024
dim_W2 = 128
dim_W3 = 64
dim_channel = 1
visualize_dim = 196
trX, vaX, teX, trY, vaY, teY = mnist_with_valid_set()
dcgan_model = DCGAN(
batch_size=batch_size,
image_shape=image_shape,
dim_z=dim_z,
dim_W1=dim_W1,
dim_W2=dim_W2,
dim_W3=dim_W3,
)
Z_tf, Y_tf, image_tf, d_cost_tf, g_cost_tf, p_real, p_gen = dcgan_model.build_model()
sess = tf.InteractiveSession()
saver = tf.train.Saver(max_to_keep=10)
discrim_vars = filter(lambda x: x.name.startswith('discrim'), tf.trainable_variables())