Skip to content

Commit

Permalink
run part 2
Browse files Browse the repository at this point in the history
@rasbt
rasbt committed Nov 5, 2019
1 parent 750dab3 commit 101d3fc
Showing 2 changed files with 353 additions and 253 deletions.
360 changes: 224 additions & 136 deletions ch17/ch17_part2.ipynb
View file

Large diffs are not rendered by default.

246 changes: 129 additions & 117 deletions ch17/ch17_part2.py
View file
Original file line number Diff line number Diff line change
@@ -15,8 +15,7 @@
#
# Code License: [MIT License](https://github.com/rasbt/python-machine-learning-book-3rd-edition/blob/master/LICENSE.txt)

# Chapter 17: Generative Adversarial Networks (part 2/2)
# =====
# # Chapter 17: Generative Adversarial Networks (Part 2/2)

# Note that the optional watermark extension is a small IPython notebook plugin that I developed to make the code reproducible. You can just skip the following line(s).

@@ -70,15 +69,19 @@



#import tensorflow as tf
#print("GPU Available: ", tf.test.is_gpu_available())
#device_name = tf.test.gpu_device_name()
#device_name


print(tf.__version__)

print("GPU Available:", tf.test.is_gpu_available())

if tf.test.is_gpu_available():
device_name = tf.test.gpu_device_name()

else:
device_name = 'CPU:0'

print(device_name)



@@ -87,11 +90,16 @@



def make_dcgan_generator(z_size=20, output_size=(28, 28, 1),
n_filters=128, n_blocks=2):
def make_dcgan_generator(
z_size=20,
output_size=(28, 28, 1),
n_filters=128,
n_blocks=2):
size_factor = 2**n_blocks
hidden_size = (output_size[0]//size_factor,
output_size[1]//size_factor)
hidden_size = (
output_size[0]//size_factor,
output_size[1]//size_factor
)

model = tf.keras.Sequential([
tf.keras.layers.Input(shape=(z_size,)),
@@ -114,21 +122,25 @@ def make_dcgan_generator(z_size=20, output_size=(28, 28, 1),
nf = n_filters
for i in range(n_blocks):
nf = nf // 2
model.add(tf.keras.layers.Conv2DTranspose(
filters=nf, kernel_size=(5, 5), strides=(2, 2),
padding='same', use_bias=False))
model.add(
tf.keras.layers.Conv2DTranspose(
filters=nf, kernel_size=(5, 5), strides=(2, 2),
padding='same', use_bias=False))
model.add(tf.keras.layers.BatchNormalization())
model.add(tf.keras.layers.LeakyReLU())


model.add(tf.keras.layers.Conv2DTranspose(
filters=output_size[2], kernel_size=(5, 5), strides=(1, 1),
padding='same', use_bias=False, activation='tanh'))
model.add(
tf.keras.layers.Conv2DTranspose(
filters=output_size[2], kernel_size=(5, 5),
strides=(1, 1), padding='same', use_bias=False,
activation='tanh'))

return model

def make_dcgan_discriminator(input_size=(28, 28, 1),
n_filters=64, n_blocks=2):
def make_dcgan_discriminator(
input_size=(28, 28, 1),
n_filters=64,
n_blocks=2):
model = tf.keras.Sequential([
tf.keras.layers.Input(shape=input_size),
tf.keras.layers.Conv2D(
@@ -141,9 +153,10 @@ def make_dcgan_discriminator(input_size=(28, 28, 1),
nf = n_filters
for i in range(n_blocks):
nf = nf*2
model.add(tf.keras.layers.Conv2D(
filters=nf, kernel_size=(5, 5),
strides=(2, 2),padding='same'))
model.add(
tf.keras.layers.Conv2D(
filters=nf, kernel_size=(5, 5),
strides=(2, 2),padding='same'))
model.add(tf.keras.layers.BatchNormalization())
model.add(tf.keras.layers.LeakyReLU())
model.add(tf.keras.layers.Dropout(0.3))
@@ -183,7 +196,6 @@ def make_dcgan_discriminator(input_size=(28, 28, 1),




mnist_bldr = tfds.builder('mnist')
mnist_bldr.download_and_prepare()
mnist = mnist_bldr.as_dataset(shuffle_files=False)
@@ -194,48 +206,28 @@ def preprocess(ex, mode='uniform'):

image = image*2 - 1.0
if mode == 'uniform':
input_z = tf.random.uniform(shape=(z_size,),
minval=-1.0, maxval=1.0)
input_z = tf.random.uniform(
shape=(z_size,), minval=-1.0, maxval=1.0)
elif mode == 'normal':
input_z = tf.random.normal(shape=(z_size,))
input_z = tf.random.normal(shape=(z_size,))
return input_z, image




num_epochs = 100
batch_size = 64
batch_size = 128
image_size = (28, 28)
z_size = 20
mode_z = 'uniform'
gen_hidden_layers = 1
gen_hidden_size = 100
disc_hidden_layers = 1
disc_hidden_size = 100
lambda_gp = 10.0

tf.random.set_seed(1)
np.random.seed(1)


if mode_z == 'uniform':
fixed_z = tf.random.uniform(
shape=(batch_size, z_size),
minval=-1, maxval=1)
elif mode_z == 'normal':
fixed_z = tf.random.normal(
shape=(batch_size, z_size))

def create_samples(g_model, input_z):
g_output = g_model(input_z, training=False)
images = tf.reshape(g_output, (batch_size, *image_size))
return (images+1)/2.0

## Set-up the dataset
mnist_trainset = mnist['train']
mnist_trainset = mnist_trainset.map(
lambda ex: preprocess(ex, mode=mode_z))

input_z, input_real = next(iter(mnist_trainset))
mnist_trainset = mnist_trainset.map(preprocess)

mnist_trainset = mnist_trainset.shuffle(10000)
mnist_trainset = mnist_trainset.batch(
@@ -245,84 +237,100 @@ def create_samples(g_model, input_z):
with tf.device(device_name):
gen_model = make_dcgan_generator()
gen_model.build(input_shape=(None, z_size))
gen_model.summary()

disc_model = make_dcgan_discriminator()
disc_model.build(input_shape=(None, np.prod(image_size)))
disc_model.summary()






## Loss function and optimizers:
loss_fn = tf.keras.losses.BinaryCrossentropy(from_logits=True)
g_optimizer = tf.keras.optimizers.Adam()
d_optimizer = tf.keras.optimizers.Adam()
## optimizers:
g_optimizer = tf.keras.optimizers.Adam(0.0002)
d_optimizer = tf.keras.optimizers.Adam(0.0002)

if mode_z == 'uniform':
fixed_z = tf.random.uniform(
shape=(batch_size, z_size),
minval=-1, maxval=1)
elif mode_z == 'normal':
fixed_z = tf.random.normal(
shape=(batch_size, z_size))

avg_epoch_losses = []
avg_d_vals = []
def create_samples(g_model, input_z):
g_output = g_model(input_z, training=False)
images = tf.reshape(g_output, (batch_size, *image_size))
return (images+1)/2.0

all_losses = []
epoch_samples = []

start_time = time.time()

for epoch in range(1, num_epochs+1):
losses = []
epoch_losses = []
for i,(input_z,input_real) in enumerate(mnist_trainset):

## Compute discriminator's real-loss and its gradients:
with tf.GradientTape() as d_tape_real:
d_logits_real = disc_model(input_real, training=True)

d_labels_real = tf.ones_like(d_logits_real)# * smoothing_factor
## Compute discriminator's loss and gradients:
with tf.GradientTape() as d_tape, tf.GradientTape() as g_tape:
g_output = gen_model(input_z, training=True)

d_critics_real = disc_model(input_real, training=True)
d_critics_fake = disc_model(g_output, training=True)

## Compute generator's loss:
g_loss = -tf.math.reduce_mean(d_critics_fake)

## Compute discriminator's losses
d_loss_real = -tf.math.reduce_mean(d_critics_real)
d_loss_fake = tf.math.reduce_mean(d_critics_fake)
d_loss = d_loss_real + d_loss_fake

## Gradient penalty:
with tf.GradientTape() as gp_tape:
alpha = tf.random.uniform(
shape=[d_critics_real.shape[0], 1, 1, 1],
minval=0.0, maxval=1.0)
interpolated = (
alpha*input_real + (1-alpha)*g_output)
gp_tape.watch(interpolated)
d_critics_intp = disc_model(interpolated)

d_loss_real = loss_fn(y_true=d_labels_real,
y_pred=d_logits_real)
d_grads_real = d_tape_real.gradient(
d_loss_real, disc_model.trainable_variables)
## Optimization: Apply the gradients
grads_intp = gp_tape.gradient(
d_critics_intp, [interpolated,])[0]
grads_intp_l2 = tf.sqrt(
tf.reduce_sum(tf.square(grads_intp), axis=[1, 2, 3]))
grad_penalty = tf.reduce_mean(tf.square(grads_intp_l2 - 1.0))

d_loss = d_loss + lambda_gp*grad_penalty

## Optimization: Compute the gradients apply them
d_grads = d_tape.gradient(d_loss, disc_model.trainable_variables)
d_optimizer.apply_gradients(
grads_and_vars=zip(d_grads_real,
disc_model.trainable_variables))

grads_and_vars=zip(d_grads, disc_model.trainable_variables))

## Compute generator's loss and its gradients:
with tf.GradientTape() as g_tape:
g_output = gen_model(input_z)
d_logits_fake = disc_model(g_output, training=True)
labels_real = tf.ones_like(d_logits_fake)
g_loss = loss_fn(y_true=labels_real,
y_pred=d_logits_fake)

g_grads = g_tape.gradient(g_loss, gen_model.trainable_variables)
g_optimizer.apply_gradients(
grads_and_vars=zip(g_grads, gen_model.trainable_variables))


## Compute discriminator's fake-loss and its gradients:
with tf.GradientTape() as d_tape_fake:
d_logits_fake = disc_model(g_output.numpy(), training=True)
d_labels_fake = tf.zeros_like(d_logits_fake)

d_loss_fake = loss_fn(y_true=d_labels_fake,
y_pred=d_logits_fake)

d_grads_fake = d_tape_fake.gradient(
d_loss_fake, disc_model.trainable_variables)
## Optimization: Apply the gradients
d_optimizer.apply_gradients(
grads_and_vars=zip(d_grads_fake,
disc_model.trainable_variables))

d_loss = (d_loss_real + d_loss_fake)/2.0
losses.append(
epoch_losses.append(
(g_loss.numpy(), d_loss.numpy(),
d_loss_real.numpy(), d_loss_fake.numpy()))


d_probs_real = tf.reduce_mean(tf.sigmoid(d_logits_real))
d_probs_fake = tf.reduce_mean(tf.sigmoid(d_logits_fake))
avg_d_vals.append((d_probs_real.numpy(), d_probs_fake.numpy()))
avg_epoch_losses.append(np.mean(losses, axis=0))

all_losses.append(epoch_losses)

print('Epoch {:-3d} | ET {:.2f} min | Avg Losses >>'
' G/D {:.4f}/{:.4f} [D-Real: {:.4f} D-Fake: {:.4f}]'
' G/D {:6.2f}/{:6.2f} [D-Real: {:6.2f} D-Fake: {:6.2f}]'
.format(epoch, (time.time() - start_time)/60,
*list(avg_epoch_losses[-1])))
epoch_samples.append(create_samples(
gen_model, num_samples=8).numpy())
*list(np.mean(all_losses[-1], axis=0)))
)

epoch_samples.append(
create_samples(gen_model, fixed_z).numpy()
)



@@ -367,7 +375,7 @@ def create_samples(g_model, input_z):
ax.tick_params(axis='both', which='major', labelsize=15)
ax2.tick_params(axis='both', which='major', labelsize=15)

#plt.savefig('/content/drive/My Drive/Colab Notebooks/PyML-3rd-edition/ch17-wdcgan-learning-curve.pdf')
#plt.savefig('images/ch17-wdcgan-learning-curve.pdf')
plt.show()


@@ -381,30 +389,34 @@ def create_samples(g_model, input_z):
ax.set_xticks([])
ax.set_yticks([])
if j == 0:
ax.text(-0.06, 0.5, 'Epoch {}'.format(e),
rotation=90, size=18, color='red',
horizontalalignment='right',
verticalalignment='center',
transform=ax.transAxes)
ax.text(
-0.06, 0.5, 'Epoch {}'.format(e),
rotation=90, size=18, color='red',
horizontalalignment='right',
verticalalignment='center',
transform=ax.transAxes)

image = epoch_samples[e-1][j]
ax.imshow(image, cmap='gray_r')

#plt.savefig('/content/drive/My Drive/Colab Notebooks/PyML-3rd-edition/ch17-wdcgan-samples.pdf')
#plt.savefig('images/ch17-wdcgan-samples.pdf')
plt.show()







# ## Mode collapse





#
# ----

#
#
# Readers may ignore the next cell.
#
#



0 comments on commit 101d3fc

Please sign in to comment.