def critic_optimizer(self):
discounted_reward = K.placeholder(shape=(None, ))
value = self.critic.output
loss = K.mean(K.square(discounted_reward - value))
optimizer = Adam(lr=self.critic_lr)
updates = optimizer.get_updates(self.critic.trainable_weights, [], loss)
train = K.function([self.critic.input, discounted_reward], [], updates=updates)
return train
def optimizer(self):
action = K.placeholder(shape=[None, 5])
discounted_rewards = K.placeholder(shape=[None, ])
# Calculate cross entropy error function
action_prob = K.sum(action * self.model.output, axis=1)
cross_entropy = K.log(action_prob) * discounted_rewards
loss = -K.sum(cross_entropy)
# create training function
optimizer = Adam(lr=self.learning_rate)
updates = optimizer.get_updates(self.model.trainable_weights, [],
loss)
train = K.function([self.model.input, action, discounted_rewards], [],
updates=updates)
return train
def actor_optimizer(self):
action = K.placeholder(shape=(None, self.action_size))
advantages = K.placeholder(shape=(None, ))
policy = self.actor.output
good_prob = K.sum(action * policy, axis=1)
eligibility = K.log(good_prob + 1e-10) * K.stop_gradient(advantages)
loss = -K.sum(eligibility)
entropy = K.sum(policy * K.log(policy + 1e-10), axis=1)
actor_loss = loss + 0.01*entropy
optimizer = Adam(lr=self.actor_lr)
updates = optimizer.get_updates(self.actor.trainable_weights, [], actor_loss)
train = K.function([self.actor.input, action, advantages], [], updates=updates)
return train
model = Model(inputs=vgg.input, outputs=predictions)
# set trainable
for layer in model.layers[:19]:
layer.trainable = False
for layer in model.layers[19:]:
layer.trainable = True
layers = [(layer, layer.name, layer.trainable) for layer in model.layers]
print("layer, layer.name, layer.trainable")
for layer in layers:
print(layer)
Epochs = 10
stepPerEpochs = int(trainSplSize / batchSize)
model.compile(Adam(lr=.003),
loss='categorical_crossentropy',
metrics=['accuracy'])
print("model.summary()")
print(model.summary())
history = model.fit_generator(train_generator,
steps_per_epoch=stepPerEpochs,
epochs=Epochs,
validation_data=val_generator,
validation_steps=50,
verbose=1)
model.save('ASL_vgg16ft_r5.h5')
# results
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(12, 4))
def gan(self):
# initialize a GAN trainer
# this is the fastest way to train a GAN in Keras
# two models are updated simutaneously in one pass
noise = Input(shape=self.generator.input_shape[1:])
real_data = Input(shape=self.discriminator.input_shape[1:])
generated = self.generator(noise)
gscore = self.discriminator(generated)
rscore = self.discriminator(real_data)
def log_eps(i):
return K.log(i+1e-11)
# single side label smoothing: replace 1.0 with 0.9
dloss = - K.mean(log_eps(1-gscore) + .1 * log_eps(1-rscore) + .9 * log_eps(rscore))
gloss = - K.mean(log_eps(gscore))
Adam = tf.train.AdamOptimizer
lr,b1 = 1e-4,.2 # otherwise won't converge.
optimizer = Adam(lr)
grad_loss_wd = optimizer.compute_gradients(dloss, self.discriminator.trainable_weights)
update_wd = optimizer.apply_gradients(grad_loss_wd)
grad_loss_wg = optimizer.compute_gradients(gloss, self.generator.trainable_weights)
update_wg = optimizer.apply_gradients(grad_loss_wg)
def get_internal_updates(model):
# get all internal update ops (like moving averages) of a model
inbound_nodes = model.inbound_nodes
input_tensors = []
for ibn in inbound_nodes:
input_tensors+= ibn.input_tensors
updates = [model.get_updates_for(i) for i in input_tensors]
return updates
other_parameter_updates = [get_internal_updates(m) for m in [self.discriminator,self.generator]]
# those updates includes batch norm.
print('other_parameter_updates for the models(mainly for batch norm):')
print(other_parameter_updates)
train_step = [update_wd, update_wg, other_parameter_updates]
losses = [dloss,gloss]
learning_phase = K.learning_phase()
def gan_feed(sess,batch_image,z_input):
# actual GAN trainer
nonlocal train_step,losses,noise,real_data,learning_phase
res = sess.run([train_step,losses],feed_dict={
noise:z_input,
real_data:batch_image,
learning_phase:True,
# Keras layers needs to know whether
# this run is training or testring (you know, batch norm and dropout)
})
loss_values = res[1]
return loss_values #[dloss,gloss]
return gan_feed
x = base_model.output
x = attach_attention_module(x, attention_module)
x = GlobalAveragePooling2D()(x)
x = Dropout(dropout)(x)
x = Dense(1024, activation='relu')(x)
x = Dropout(dropout)(x)
predictions = Dense(no_of_classes, activation='softmax')(x)
lr_scheduler = LearningRateScheduler(lr_schedule)
lr_reducer = ReduceLROnPlateau(factor=np.sqrt(0.1),
cooldown=0,
patience=2,
min_lr=0.5e-6)
model = Model(input=base_model.input, output=predictions)
model.compile(optimizer=Adam(lr_schedule(0)),
loss='categorical_crossentropy',
metrics=['categorical_accuracy', 'accuracy'])
filepath = output_models_dir + experiment_name + "_{epoch:02d}_{val_acc:.2f}.h5"
checkpoint = ModelCheckpoint(filepath,
monitor='val_loss',
verbose=1,
save_best_only=False,
save_weights_only=False,
mode='auto',
period=1)
checkpoints = [checkpoint]
model.fit_generator(train_generator,
epochs=epochs,
steps_per_epoch=87,
#model.add(layers.UpSampling2D((2,2)))
#model.add(layers.UpSampling2D((2,2)))
model.add(base_model)
model.add(layers.Flatten())
#model.add(layers.BatchNormalization())
#model.add(layers.Dense(128, activation='relu'))
#model.add(layers.Dropout(0.5))
#model.add(layers.BatchNormalization())
#model.add(layers.Dense(64, activation='relu'))
#model.add(layers.Dropout(0.5))
#model.add(layers.BatchNormalization())
model.add(layers.Dense(
10, activation='softmax')) #, kernel_initializer='he_uniform'))
model.compile(loss='categorical_crossentropy',
optimizer=Adam(lr=args_lr),
metrics=['accuracy'])
#model.summary()
print(model_type)
#pdb.set_trace()
current_epoch = 0
################### connects interrupt signal to the process #####################
def terminateProcess(signalNumber, frame):
print('checkpointing the model triggered by kill -15 signal')
# delete whatever checkpoint that already exists
def model_train(model, batch_size=32, epochs=200, data_path = None, data_augmentation = True, subtract_pixel_mean = True):
num_classes = 10
(x_train, y_train), (x_test, y_test) = load_data(data_path)
x_train = x_train.astype('float32') / 255
x_test = x_test.astype('float32') / 255
# If subtract pixel mean is enabled
if subtract_pixel_mean:
x_train_mean = np.mean(x_train, axis=0)
x_train -= x_train_mean
x_test -= x_train_mean
# Convert class vectors to binary class matrices.
y_train = np_utils.to_categorical(y_train, num_classes)
y_test = np_utils.to_categorical(y_test, num_classes)
model.compile(loss='categorical_crossentropy',
optimizer=Adam(lr=lr_schedule(0)),
metrics=['accuracy'])
checkpoint = ModelCheckpoint(filepath='./weights/resnet20_cifar10_weights.{epoch:03d}.h5',
monitor='loss',
save_best_only=True,
save_weights_only=True)
lr_scheduler = LearningRateScheduler(lr_schedule)
lr_reducer = ReduceLROnPlateau(factor=np.sqrt(0.1),
cooldown=0,
patience=5,
min_lr=0.5e-6)
csv_logger = CSVLogger('./results/training_resnet20_cifar10.csv')
callbacks = [checkpoint, lr_reducer, lr_scheduler, csv_logger]
if not data_augmentation:
print('Not using data augmentation.')
model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
validation_data=(x_test, y_test),
shuffle=True,
callbacks=callbacks)
else:
print('Using real-time data augmentation.')
# This will do preprocessing and realtime data augmentation:
datagen = ImageDataGenerator(featurewise_center=False,# set input mean to 0 over the dataset
samplewise_center=False,# set each sample mean to 0
featurewise_std_normalization=False,# divide inputs by std of dataset
samplewise_std_normalization=False,# divide each input by its std
zca_whitening=False,# apply ZCA whitening
rotation_range=0,# randomly rotate images in the range (deg 0 to 180)
width_shift_range=0.1,# randomly shift images horizontally
height_shift_range=0.1,# randomly shift images vertically
horizontal_flip=True,# randomly flip images
vertical_flip=False)# randomly flip images
# Compute quantities required for featurewise normalization
# (std, mean, and principal components if ZCA whitening is applied).
datagen.fit(x_train)
# Fit the model on the batches generated by datagen.flow().
model.fit_generator(datagen.flow(x_train, y_train, batch_size=batch_size),
validation_data=(x_test, y_test),
epochs=epochs,
verbose=1,
workers=8,
use_multiprocessing=True,
callbacks=callbacks)
def main():
dataList = []
#plt.imshow(preprocessImage(img, transform))
#plt.show()
#showSamplesCompared(img, transform, '', '', '')
#plt.xkcd()
np.random.seed(0)
#data = pd.read_csv('/home/jjordening/data/dataset_sim_000_km_few_laps/driving_log.csv',
# header = None, names=['center','left', 'right', 'steering','throttle', 'brake', 'speed', 'position', 'orientation'])
#data['positionX'], data['positionY'], data['positionZ'] = data['position'].apply(retrieveVectors)
#data['orientationX'], data['orientationY'], data['orientationZ'] = data['orientation'].apply(retrieveVectors)
#data['center'] = '/home/jjordening/data/dataset_sim_000_km_few_laps/'+data['center'].apply(lambda x: x.strip())
#data1 = pd.read_csv('/home/jjordening/data/udacity-day-01-exported-1102/output_processed.txt')
#data1['path'] = '/home/jjordening/data/udacity-day-01-exported-1102/'+data1['path'].apply(lambda x: x.strip())
#data2 = pd.read_csv('/home/jjordening/data/udacity-day-01-exported-1109/output_processed.txt')
#data2['path'] = '/home/jjordening/data/udacity-day-01-exported-1109/'+data2['path'].apply(lambda x: x.strip())
if ALL:
data3 = pd.read_csv('/home/jjordening/data/1538/output_processed.txt')
data3['path'] = '/home/jjordening/data/1538/' + data3['path'].apply(
lambda x: x.strip())
print('data3', np.max(data3['steering']), np.min(data3['steering']))
dataList.append(data3)
data4 = pd.read_csv('/home/jjordening/data/1543/output_processed.txt')
data4['path'] = '/home/jjordening/data/1543/' + data4['path'].apply(
lambda x: x.strip())
print('data4', np.max(data4['steering']), np.min(data4['steering']))
dataList.append(data4)
data5 = pd.read_csv('/home/jjordening/data/1610/output_processed.txt')
data5['path'] = '/home/jjordening/data/1610/' + data5['path'].apply(
lambda x: x.strip())
print('data5', np.max(data5['steering']), np.min(data5['steering']))
dataList.append(data5)
data6 = pd.read_csv('/home/jjordening/data/1645/output_processed.txt')
data6['path'] = '/home/jjordening/data/1645/' + data6['path'].apply(
lambda x: x.strip())
print('data6', np.max(data6['steering']), np.min(data6['steering']))
dataList.append(data6)
data7 = pd.read_csv('/home/jjordening/data/1702/output_processed.txt')
data7['path'] = '/home/jjordening/data/1702/' + data7['path'].apply(
lambda x: x.strip())
print('data7', np.max(data7['steering']), np.min(data7['steering']))
dataList.append(data7)
data8 = pd.read_csv('/home/jjordening/data/1708/output_processed.txt')
data8['path'] = '/home/jjordening/data/1708/' + data8['path'].apply(
lambda x: x.strip())
print('data8', np.max(data8['steering']), np.min(data8['steering']))
dataList.append(data8)
data9 = pd.read_csv('/home/jjordening/data/1045/output_processed.txt')
data9['path'] = '/home/jjordening/data/1045/' + data9['path'].apply(
lambda x: x.strip())
print('data9', np.max(data9['steering']), np.min(data9['steering']))
dataList.append(data9)
data10 = pd.read_csv('/home/jjordening/data/1050/output_processed.txt')
data10['path'] = '/home/jjordening/data/1050/' + data10['path'].apply(
lambda x: x.strip())
print('data10', np.max(data10['steering']), np.min(data10['steering']))
dataList.append(data10)
data11 = pd.read_csv('/home/jjordening/data/1426/output_processed.txt')
data11['path'] = '/home/jjordening/data/1426/' + data11['path'].apply(
lambda x: x.strip())
print('data11', np.max(data11['steering']), np.min(data11['steering']))
dataList.append(data11)
data12 = pd.read_csv('/home/jjordening/data/1516/output_processed.txt')
data12['path'] = '/home/jjordening/data/1516/' + data12['path'].apply(
lambda x: x.strip())
print('data12', np.max(data12['steering']), np.min(data12['steering']))
dataList.append(data12)
data13 = pd.read_csv('/home/jjordening/data/1634/outputNew.txt')
data13['path'] = '/home/jjordening/data/1634/' + data13['path'].apply(
lambda x: x.strip())
print('data12', np.max(data13['steering']), np.min(data13['steering']))
dataList.append(data13)
print(data9['brake'].unique())
"""data3 = pd.read_csv('/home/jjordening/data/dataset_polysync_1464552951979919/output_processed.txt', header = None,
names = ['path','heading','longitude','latitude','quarternion0','quarternion1','quarternion2','quarternion3','vel0','vel1',
'vel2','steering','throttle','brake','speed'], skiprows = 500)
data3 = data3.ix[0:1500].append(data3.ix[2600:])
data3 = data3.ix[-500:]
data3['path'] = '/home/jjordening/data/dataset_polysync_1464552951979919/'+data3['path'].apply(lambda x: x.strip())
data3['throttle'] = 0"""
#data['right'] = '../simulator/data/data/'+data['right'].apply(lambda x: x.strip())
#data['left'] = '../simulator/data/data/'+data['left'].apply(lambda x: x.strip())
angles = []
dataNew = pd.DataFrame()
offset = 0
#print(data3['steering'])
#print(data1['longitude'])
"""for dat in [data3,data4,data5,data6,data7]:
angles.extend(dat['steering'].values)
for row in dat.iterrows():
dat.loc[row[0], 'angleIndex'] = row[0]+ offset
#images.append(preprocessImage(mpimg.imread(row[1]['center'].strip())))
#images.append(transform(mpimg.imread(row[1]['center'].strip())))
offset+=100
dataNew = dataNew.append(dat.ix[100:])"""
#dataNew['throttle'] = dataNew['accel'].apply(lambda x: max(x,0)/np.max(dataNew['accel']))
for dat in dataList:
dataNew = dataNew.append(dat.ix[30:])
del dat
print('Len dataNew: ', len(dataNew))
dataNew = dataNew.loc[pd.notnull(dataNew['throttle'])]
dataNew = dataNew.loc[pd.notnull(dataNew['brake'])]
dataNew = dataNew.loc[pd.notnull(dataNew['steering'])]
print('Len dataNew: ', len(dataNew))
print(np.max(dataNew['throttle']), np.min(dataNew['throttle']))
# TODO: Normalisation of position and orientation<
#del data3,data4,data5,data6,data7
print(len(dataNew), dataNew.columns)
print(np.histogram(dataNew['throttle'], bins=31))
hist, edges = np.histogram(dataNew['steering'], bins=31)
print(hist, edges, len(dataNew))
hist = 1. / np.array([
val if val > len(dataNew) / 40. else len(dataNew) / 40. for val in hist
])
hist *= len(dataNew) / 40.
print(hist, edges, len(dataNew))
dataNew['norm'] = dataNew['steering'].apply(
lambda x: getNormFactor(x, hist, edges))
print(dataNew['norm'].unique())
print(np.min(dataNew['steering']), np.max(dataNew['steering']))
print(np.min(dataNew['throttle']), np.max(dataNew['throttle']))
print(np.min(dataNew['brake']), np.max(dataNew['brake']))
dataNew['speed'] = dataNew['speed'].apply(lambda x: x / 40. - 1)
dataNew = shuffle(dataNew, random_state=0)
#plt.figure(1, figsize=(8,4))
#plt.hist(dataNew['steering'], bins =31)
#plt.show()
dataTrain, dataTest = train_test_split(dataNew, test_size=.1)
dataTrain, dataVal = train_test_split(dataTrain, test_size=.1)
file = open(dataTrain['path'].iloc[0], 'rb')
# Use the PIL raw decoder to read the data.
# - the 'F;16' informs the raw decoder that we are reading a little endian, unsigned integer 16 bit data.
img = np.array(Image.frombytes('RGB', [960, 480], file.read(), 'raw'))
file.close()
imShape = preprocessImage(img).shape
print(imShape)
batchSize = 128
epochBatchSize = 8192
trainGenerator = generateImagesFromPaths(dataTrain, batchSize, imShape,
[3], True)
t = time.time()
trainGenerator.__next__()
print("Time to build train batch: ", time.time() - t)
valGenerator = generateImagesFromPaths(dataVal, batchSize, imShape, [3])
t = time.time()
valGenerator.__next__()
print("Time to build validation batch: ", time.time() - t)
stopCallback = EarlyStopping(monitor='val_loss',
patience=20,
min_delta=0.01)
checkCallback = ModelCheckpoint('psyncModel.ckpt',
monitor='val_loss',
save_best_only=True)
visCallback = TensorBoard(log_dir='./logs/%d' % int(time.time()),
histogram_freq=0,
write_graph=True,
write_images=True)
model = load_model('psyncModelBase.h5',
custom_objects={'customLoss': customLoss})
name = 'psyncPosNet%d.h5' % time.time()
if LOADMODEL:
endModel = load_model('psyncPosNet.h5',
custom_objects={'customLoss': customLoss})
endModel.fit_generator(
trainGenerator,
callbacks=[stopCallback, checkCallback, visCallback],
nb_epoch=100,
samples_per_epoch=epochBatchSize,
max_q_size=8,
validation_data=valGenerator,
nb_val_samples=len(dataVal),
nb_worker=8,
pickle_safe=True)
endModel.load_weights('psyncModel.ckpt')
endModel.save(name)
else:
inpC = Input(shape=(imShape[0], imShape[1], imShape[2]),
name='inputImg')
xC = Convolution2D(24,
8,
8,
border_mode='valid',
subsample=(2, 2),
name='conv1')(inpC)
xC = BatchNormalization()(xC)
xC = Activation('elu')(xC)
print(xC.get_shape())
xC = Convolution2D(36,
5,
5,
border_mode='valid',
subsample=(2, 2),
name='conv2')(xC)
xC = BatchNormalization()(xC)
xC = Activation('elu')(xC)
print(xC.get_shape())
xC = Convolution2D(48,
5,
5,
border_mode='valid',
subsample=(2, 2),
name='conv3')(xC)
xC = BatchNormalization()(xC)
xC = Activation('elu')(xC)
print(xC.get_shape())
xC = Convolution2D(64, 5, 5, border_mode='valid', name='conv4')(xC)
xC = BatchNormalization()(xC)
xC = Activation('elu')(xC)
print(xC.get_shape())
xC = Convolution2D(
64,
5,
5,
border_mode='valid',
name='conv5',
)(xC)
xC = BatchNormalization()(xC)
xC = Activation('elu')(xC)
print(xC.get_shape())
xOut = Flatten()(xC)
print(xOut.get_shape())
#Cut for transfer learning is here:
speedInp = Input(shape=(1, ), name='inputSpeed')
xOut = Lambda(lambda x: K.concatenate(x, axis=1))([xOut, speedInp])
xOut = Dense(200)(xOut)
xOut = BatchNormalization()(xOut)
xOut = Activation('elu')(xOut)
xOut = Dense(200)(xOut)
xOut = BatchNormalization()(xOut)
xEnd = Activation('elu')(xOut)
xOutSteer = Dense(100)(xEnd)
xOutSteer = BatchNormalization()(xOutSteer)
xOutSteer = Activation('elu')(xOutSteer)
xOutSteer = Dropout(.2)(xOutSteer)
xOutSteer = Dense(30)(xOutSteer)
xOutSteer = BatchNormalization()(xOutSteer)
xOutSteer = Activation('elu')(xOutSteer)
xOutSteer = Dense(1, activation='sigmoid')(xOutSteer)
xOutSteer = Lambda(lambda x: x * 10 - 5, name='outputSteer')(xOutSteer)
xOutThr = Dense(100, name='thr1')(xEnd)
xOutThr = BatchNormalization(name='thr2')(xOutThr)
xOutThr = Activation('elu')(xOutThr)
xOutThr = Dropout(.2)(xOutThr)
xOutThr = Dense(30, name='thr3')(xOutThr)
xOutThr = BatchNormalization(name='thr4')(xOutThr)
xOutThr = Activation('elu')(xOutThr)
xOutThr = Dense(1, activation='sigmoid', name='thr5')(xOutThr)
xOutThr = Lambda(lambda x: x * 2 - 1, name='outputThr')(xOutThr)
xOutPos = Dropout(.3)(xEnd)
xOutPos = Dense(1, activation='sigmoid', name='pos5')(xOutPos)
xOutPos = Lambda(lambda x: x * 2 - 1, name='outputPos')(xOutPos)
endModel = Model((inpC, speedInp), (xOutSteer, xOutThr, xOutPos))
endModel.compile(optimizer=Adam(lr=1e-4), loss='mse', metrics=['mse'])
#endModel.fit_generator(trainGenerator, callbacks = [visCallback],
# nb_epoch=50, samples_per_epoch=epochBatchSize,
# max_q_size=24, nb_worker=8, pickle_safe=True)
endModel.fit_generator(
trainGenerator,
callbacks=[stopCallback, checkCallback, visCallback],
nb_epoch=30,
samples_per_epoch=epochBatchSize,
nb_worker=8,
pickle_safe=True)
endModel.load_weights('psyncModel.ckpt')
endModel.save(name)
endModel.fit_generator(
trainGenerator,
callbacks=[stopCallback, checkCallback, visCallback],
nb_epoch=30,
samples_per_epoch=epochBatchSize,
nb_worker=8,
pickle_safe=True)
endModel.load_weights('psyncModel.ckpt')
endModel.save(name)
endModel.fit_generator(
trainGenerator,
callbacks=[stopCallback, checkCallback, visCallback],
nb_epoch=40,
samples_per_epoch=epochBatchSize,
max_q_size=24,
validation_data=valGenerator,
nb_val_samples=len(dataVal),
nb_worker=8,
pickle_safe=True)
endModel.load_weights('psyncModel.ckpt')
endModel.save(name)
endModel = load_model(name, custom_objects={'customLoss': customLoss})
print(endModel.evaluate_generator(valGenerator, val_samples=len(dataVal)))
print(
endModel.evaluate_generator(generateImagesFromPaths(
dataTest, batchSize, imShape, [3], angles),
val_samples=len(dataTest)))
outputs.append(out)
# Step 3: Create model instance taking three inputs and returning the list of outputs. (≈ 1 line)
model = Model(inputs=[X, s0, c0], outputs=outputs)
### END CODE HERE ###
return model
#Create the model
model = model(Tx, Ty, n_a, n_s, len(human_vocab), len(machine_vocab))
model.summary()
opt = Adam()
model.compile(optimizer=opt,
loss="categorical_crossentropy",
metrics=["accuracy"])
s0 = np.zeros((m, n_s))
c0 = np.zeros((m, n_s))
outputs = list(Yoh.swapaxes(0, 1))
model.fit([Xoh, s0, c0], outputs, epochs=1, batch_size=100)
model.load_weights('models/model.h5')
EXAMPLES = [
'3 May 1979', '5 April 09', '21th of August 2016', 'Tue 10 Jul 2007',
'Saturday May 9 2018', 'March 3 2001', 'March 3rd 2001', '1 March 2001'
]
def get_net():
inputs = Input(shape=(img_h, img_w, N_channels))
# 网络结构定义
conv1 = Conv2D(64,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(inputs)
conv1 = Conv2D(64,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(128,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(pool1)
conv2 = Conv2D(128,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv2)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
conv3 = Conv2D(256,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(pool2)
conv3 = Conv2D(256,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv3)
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
conv4 = Conv2D(512,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(pool3)
conv4 = Conv2D(512,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv4)
drop4 = Dropout(0.5)(conv4)
pool4 = MaxPooling2D(pool_size=(2, 2))(drop4)
conv5 = Conv2D(1024,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(pool4)
conv5 = Conv2D(1024,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv5)
drop5 = Dropout(0.5)(conv5)
up6 = Conv2D(512,
2,
activation='relu',
padding='same',
kernel_initializer='he_normal')(UpSampling2D(size=(2,
2))(drop5))
merge6 = concatenate([drop4, up6], axis=3)
conv6 = Conv2D(512,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(merge6)
conv6 = Conv2D(512,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv6)
up7 = Conv2D(256,
2,
activation='relu',
padding='same',
kernel_initializer='he_normal')(UpSampling2D(size=(2,
2))(conv6))
merge7 = concatenate([conv3, up7], axis=3)
conv7 = Conv2D(256,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(merge7)
conv7 = Conv2D(256,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv7)
up8 = Conv2D(128,
2,
activation='relu',
padding='same',
kernel_initializer='he_normal')(UpSampling2D(size=(2,
2))(conv7))
merge8 = concatenate([conv2, up8], axis=3)
conv8 = Conv2D(128,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(merge8)
conv8 = Conv2D(128,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv8)
up9 = Conv2D(64,
2,
activation='relu',
padding='same',
kernel_initializer='he_normal')(UpSampling2D(size=(2,
2))(conv8))
merge9 = concatenate([conv1, up9], axis=3)
conv9 = Conv2D(64,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(merge9)
conv9 = Conv2D(64,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv9)
conv9 = Conv2D(2,
3,
activation='relu',
padding='same',
kernel_initializer='he_normal')(conv9)
conv10 = Conv2D(C, (1, 1),
activation='relu',
kernel_initializer='he_normal')(conv9)
reshape = Reshape((C, img_h * img_w),
input_shape=(C, img_h, img_w))(conv10)
reshape = Permute((2, 1))(reshape)
activation = Activation('softmax')(reshape)
model = Model(input=inputs, output=activation)
model.compile(optimizer=Adam(lr=1.0e-4),
loss='categorical_crossentropy',
metrics=['accuracy'])
return model
def training_game():
env = Environment(
map_name="ForceField",
visualize=True,
game_steps_per_episode=150,
agent_interface_format=features.AgentInterfaceFormat(
feature_dimensions=features.Dimensions(screen=64, minimap=32)))
input_shape = (_SIZE, _SIZE, 1)
nb_actions = 12 # Number of actions
model = neural_network_model(input_shape, nb_actions)
memory = SequentialMemory(limit=5000, window_length=_WINDOW_LENGTH)
processor = SC2Proc()
# Policy
policy = LinearAnnealedPolicy(EpsGreedyQPolicy(),
attr="eps",
value_max=1,
value_min=0.2,
value_test=.0,
nb_steps=1e2)
# Agent
dqn = DQNAgent(
model=model,
nb_actions=nb_actions,
memory=memory,
enable_double_dqn=True,
enable_dueling_network=True,
# 2019-07-12 GU Zhan (Sam)
# nb_steps_warmup=500, target_model_update=1e-2, policy=policy,
nb_steps_warmup=2000,
target_model_update=1e-2,
policy=policy,
batch_size=150,
processor=processor,
delta_clip=1)
dqn.compile(Adam(lr=.001), metrics=["mae", "acc"])
# Tensorboard callback
timestamp = f"{datetime.datetime.now():%Y-%m-%d %I:%M%p}"
# 2019-07-12 GU Zhan (Sam) folder name for Lunux:
# callbacks = keras.callbacks.TensorBoard(log_dir='./Graph/'+ timestamp, histogram_freq=0,
# write_graph=True, write_images=False)
# 2019-07-12 GU Zhan (Sam) folder name for Windows:
callbacks = keras.callbacks.TensorBoard(log_dir='.\Graph\issgz',
histogram_freq=0,
write_graph=True,
write_images=False)
# Save the parameters and upload them when needed
name = "agent"
w_file = "dqn_{}_weights.h5f".format(name)
check_w_file = "train_w" + name + "_weights.h5f"
if SAVE_MODEL:
check_w_file = "train_w" + name + "_weights_{step}.h5f"
log_file = "training_w_{}_log.json".format(name)
if LOAD_MODEL:
dqn.load_weights(w_file)
class Saver(Callback):
def on_episode_end(self, episode, logs={}):
if episode % 200 == 0:
self.model.save_weights(w_file, overwrite=True)
s = Saver()
logs = FileLogger('DQN_Agent_log.csv', interval=1)
# dqn.fit(env, callbacks=[callbacks,s,logs], nb_steps=600, action_repetition=2,
dqn.fit(env,
callbacks=[callbacks, s, logs],
nb_steps=10000,
action_repetition=2,
log_interval=1e4,
verbose=2)
dqn.save_weights(w_file, overwrite=True)
dqn.test(env, action_repetition=2, nb_episodes=30, visualize=False)
aspect_ratios_per_layer=None,
two_boxes_for_ar1=two_boxes_for_ar1,
limit_boxes=limit_boxes,
variances=variances,
coords=coords,
normalize_coords=normalize_coords)
model.load_weights('./ssd7_0_weights.h5')
#model = load_model('./ssd7_0.h5')
### Set up training
batch_size = 4
# 3: Instantiate an Adam optimizer and the SSD loss function and compile the model
adam = Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=5e-05)
ssd_loss = SSDLoss(neg_pos_ratio=3, n_neg_min=0, alpha=1.0)
model.compile(optimizer=adam, loss=ssd_loss.compute_loss)
# 4: Instantiate an encoder that can encode ground truth labels into the format needed by the SSD loss function
ssd_box_encoder = SSDBoxEncoder(img_height=img_height,
img_width=img_width,
n_classes=n_classes,
predictor_sizes=predictor_sizes,
min_scale=min_scale,
max_scale=max_scale,
scales=scales,
aspect_ratios_global=aspect_ratios,
def multihead_seq_model(tasks,
filters,
n_dil_layers,
conv1_kernel_size,
tconv_kernel_size,
b_loss_weight=1,
c_loss_weight=1,
p_loss_weight=1,
c_splines=20,
p_splines=0,
merge_profile_reg=False,
lr=0.004,
padding='same',
batchnorm=False,
use_bias=False,
n_profile_bias_tracks=2,
n_bias_tracks=2,
seqlen=None,
skip_type='residual'):
from basepair.seqmodel import SeqModel
from basepair.layers import DilatedConv1D, DeConv1D, GlobalAvgPoolFCN
from basepair.metrics import BPNetMetricSingleProfile
from basepair.heads import ScalarHead, ProfileHead
from gin_train.metrics import ClassificationMetrics, RegressionMetrics
from basepair.losses import mc_multinomial_nll_2, CountsMultinomialNLL
from basepair.exp.paper.config import peak_pred_metric
from basepair.activations import clipped_exp
from basepair.functions import softmax
assert p_loss_weight >= 0
assert c_loss_weight >= 0
assert b_loss_weight >= 0
# Heads -------------------------------------------------
heads = []
# Profile prediction
if p_loss_weight > 0:
if not merge_profile_reg:
heads.append(
ProfileHead(target_name='{task}/profile',
net=DeConv1D(n_tasks=2,
filters=filters,
tconv_kernel_size=tconv_kernel_size,
padding=padding,
n_hidden=0,
batchnorm=batchnorm),
loss=mc_multinomial_nll_2,
loss_weight=p_loss_weight,
postproc_fn=softmax,
use_bias=use_bias,
bias_input='bias/{task}/profile',
bias_shape=(None, n_profile_bias_tracks),
metric=peak_pred_metric))
else:
heads.append(
ProfileHead(
target_name='{task}/profile',
net=DeConv1D(
n_tasks=2,
filters=filters,
tconv_kernel_size=tconv_kernel_size,
padding=padding,
n_hidden=1, # use 1 hidden layer in that case
batchnorm=batchnorm),
activation=clipped_exp,
loss=CountsMultinomialNLL(2, c_task_weight=c_loss_weight),
loss_weight=p_loss_weight,
bias_input='bias/{task}/profile',
use_bias=use_bias,
bias_shape=(None, n_profile_bias_tracks),
metric=BPNetMetricSingleProfile(
count_metric=RegressionMetrics(),
profile_metric=peak_pred_metric)))
c_loss_weight = 0 # don't need to use the other count loss
# Count regression
if c_loss_weight > 0:
heads.append(
ScalarHead(
target_name='{task}/counts',
net=GlobalAvgPoolFCN(n_tasks=2,
n_splines=p_splines,
batchnorm=batchnorm),
activation=None,
loss='mse',
loss_weight=c_loss_weight,
bias_input='bias/{task}/counts',
use_bias=use_bias,
bias_shape=(n_bias_tracks, ),
metric=RegressionMetrics(),
))
# Binary classification
if b_loss_weight > 0:
heads.append(
ScalarHead(
target_name='{task}/class',
net=GlobalAvgPoolFCN(n_tasks=1,
n_splines=c_splines,
batchnorm=batchnorm),
activation='sigmoid',
loss='binary_crossentropy',
loss_weight=b_loss_weight,
metric=ClassificationMetrics(),
))
# -------------------------------------------------
m = SeqModel(
body=DilatedConv1D(filters=filters,
conv1_kernel_size=conv1_kernel_size,
n_dil_layers=n_dil_layers,
padding=padding,
batchnorm=batchnorm,
skip_type=skip_type),
heads=heads,
tasks=tasks,
optimizer=Adam(lr=lr),
seqlen=seqlen,
)
return m
def __init__(self,
numOfClasses,
batch_size=32,
grayScale=True,
leaky=False,
swish=False,
attention_middle=False,
attention_last=False,
conditional=False):
self.img_rows = 64
self.img_cols = 64
if grayScale == True:
self.channels = 1
else:
self.channels = 3
self.img_shape = (self.img_rows, self.img_cols, self.channels)
#self.latent_dim = 100
self.num_classes = numOfClasses
self.latent_dim = 80
self.losslog = []
self.batch_size = batch_size
self.grayScale = grayScale
self.leaky = leaky
self.swish = swish
self.attention_middle = attention_middle
self.attention_last = attention_last
self.conditional = conditional
self.imagPath = "images"
# Following parameter and optimizer set as recommended in paper
self.n_critic = 2 #original 2
if self.leaky:
optimizer_gen = Adam(lr=5e-05,
beta_1=0.0,
beta_2=0.7,
epsilon=1e-08) # working
optimizer_cri = Adam(lr=2e-04,
beta_1=0.0,
beta_2=0.7,
epsilon=1e-08) # working
elif self.swish:
optimizer_gen = Adam(lr=5e-05,
beta_1=0.0,
beta_2=0.7,
epsilon=1e-08) # working
optimizer_cri = Adam(lr=2e-04,
beta_1=0.0,
beta_2=0.7,
epsilon=1e-08) # working
else:
optimizer_gen = Adam(lr=5e-05,
beta_1=0.0,
beta_2=0.7,
epsilon=1e-08) # working
optimizer_cri = Adam(lr=2e-04,
beta_1=0.0,
beta_2=0.7,
epsilon=1e-08) # working
# Build the generator and critic
self.generator = self.build_generator()
self.critic = self.build_critic()
#keras2ascii(self.generator)
#plot_model(self.generator, to_file='generator_model.png')
#keras2ascii(self.critic)
#plot_model(self.critic, to_file='generator_model.png')
#-------------------------------
# Construct Computational Graph
# for the Critic
#-------------------------------
# Freeze generator's layers while training critic
self.generator.trainable = False
# Image input (real sample)
real_img = Input(shape=self.img_shape)
# Noise input
z_disc = Input(shape=(self.latent_dim, ))
# ADDING LABEL TO MAKE IT DWGAN
if self.conditional:
label = Input(shape=(1, ))
# Generate image based of noise (fake sample)
fake_img = self.generator([z_disc, label])
# Discriminator determines validity of the real and fake images
fake = self.critic([fake_img, label])
valid = self.critic([real_img, label])
# Construct weighted average between real and fake images
interpolated_img = RandomWeightedAverage()([real_img, fake_img])
# Determine validity of weighted sample
validity_interpolated = self.critic([interpolated_img, label])
# Use Python partial to provide loss function with additional
# 'averaged_samples' argument
partial_gp_loss = partial(self.gradient_penalty_loss,
averaged_samples=interpolated_img)
partial_gp_loss.__name__ = 'gradient_penalty' # Keras requires function names
self.critic_model = Model(
inputs=[real_img, label, z_disc],
outputs=[valid, fake, validity_interpolated])
self.critic_model.compile(loss=[
self.wasserstein_loss, self.wasserstein_loss, partial_gp_loss
],
optimizer=optimizer_cri,
loss_weights=[1, 1, 10])
else:
fake_img = self.generator(z_disc)
# Discriminator determines validity of the real and fake images
fake = self.critic(fake_img)
valid = self.critic(real_img)
# Construct weighted average between real and fake images
interpolated_img = RandomWeightedAverage()([real_img, fake_img])
# Determine validity of weighted sample
validity_interpolated = self.critic(interpolated_img)
# Use Python partial to provide loss function with additional
# 'averaged_samples' argument
partial_gp_loss = partial(self.gradient_penalty_loss,
averaged_samples=interpolated_img)
partial_gp_loss.__name__ = 'gradient_penalty' # Keras requires function names
self.critic_model = Model(
inputs=[real_img, z_disc],
outputs=[valid, fake, validity_interpolated])
self.critic_model.compile(loss=[
self.wasserstein_loss, self.wasserstein_loss, partial_gp_loss
],
optimizer=optimizer_cri,
loss_weights=[1, 1, 10])
#-------------------------------
# Construct Computational Graph
# for Generator
#-------------------------------
# For the generator we freeze the critic's layers
self.critic.trainable = False
self.generator.trainable = True
# Sampled noise for input to generator
z_gen = Input(shape=(self.latent_dim, ))
if self.conditional:
# add label to the input
label = Input(shape=(1, ))
# Generate images based of noise
#img = self.generator(z_gen)
img = self.generator([z_gen, label])
# Discriminator determines validity
valid = self.critic([img, label])
# Defines generator model
self.generator_model = Model([z_gen, label], valid)
else:
# Generate images based of noise
img = self.generator(z_gen)
# Discriminator determines validity
valid = self.critic(img)
# Defines generator model
self.generator_model = Model(z_gen, valid)
self.generator_model.compile(loss=self.wasserstein_loss,
optimizer=optimizer_gen)
x_fake = g_model(z_in)
x_inter = Lambda(interpolating)([x_in, x_fake])
x_real_score = d_model(x_in)
x_fake_score = d_model(x_fake)
x_inter_score = d_model(x_inter)
grads = K.gradients(x_inter_score, [x_inter])[0]
grad_norms = K.sqrt(K.sum(grads**2, range(1, K.ndim(grads))) + 1e-9)
d_train_model = Model([x_in, z_in],
[x_real_score, x_fake_score, x_inter_score])
d_loss = K.mean(-(x_real_score - x_fake_score)) + 10 * K.mean(
(grad_norms - 1)**2)
d_train_model.add_loss(d_loss)
d_train_model.compile(optimizer=Adam(2e-4, 0.5))
# 整合模型（训练生成器）
g_model.trainable = True
d_model.trainable = False
x_fake_score = d_model(g_model(z_in))
g_train_model = Model(z_in, x_fake_score)
g_train_model.add_loss(K.mean(-x_fake_score))
g_train_model.compile(optimizer=Adam(2e-4, 0.5))
# 检查模型结构
d_train_model.summary()
g_train_model.summary()
env.seed(123)
nb_actions = env.action_space.n
model = Sequential()
model.add(Flatten(input_shape=(1, ) + env.observation_space.shape))
model.add(Dense(16))
model.add(Activation('relu'))
model.add(Dense(16))
model.add(Activation('relu'))
model.add(Dense(16))
model.add(Activation('relu'))
model.add(Dense(nb_actions))
model.add(Activation('linear'))
print(model.summary())
memory = SequentialMemory(limit=50000, window_length=1)
policy = BoltzmannQPolicy()
dqn = DQNAgent(model=model,
nb_actions=nb_actions,
memory=memory,
nb_steps_warmup=10,
target_model_update=1e-2,
policy=policy)
dqn.compile(Adam(lr=1e-3), metrics=['mae'])
dqn.fit(env, nb_steps=1000, visualize=True, verbose=2)
dqn.save_weights('dqn_{}_weights.h5f'.format(ENV_NAME), overwrite=True)
dqn.test(env, nb_episodes=5, visualize=True)
def unet_model_3d(input_shape,
pool_size=(2, 2, 2),
n_labels=1,
initial_learning_rate=0.00001,
deconvolution=False,
depth=4,
n_base_filters=32,
include_label_wise_dice_coefficients=False,
metrics=dice_coefficient,
batch_normalization=False,
activation_name="sigmoid"):
"""
Builds the 3D UNet Keras model.f
:param metrics: List metrics to be calculated during model training (default is dice coefficient).
:param include_label_wise_dice_coefficients: If True and n_labels is greater than 1, model will report the dice
coefficient for each label as metric.
:param n_base_filters: The number of filters that the first layer in the convolution network will have. Following
layers will contain a multiple of this number. Lowering this number will likely reduce the amount of memory required
to train the model.
:param depth: indicates the depth of the U-shape for the model. The greater the depth, the more max pooling
layers will be added to the model. Lowering the depth may reduce the amount of memory required for training.
:param input_shape: Shape of the input data (n_chanels, x_size, y_size, z_size). The x, y, and z sizes must be
divisible by the pool size to the power of the depth of the UNet, that is pool_size^depth.
:param pool_size: Pool size for the max pooling operations.
:param n_labels: Number of binary labels that the model is learning.
:param initial_learning_rate: Initial learning rate for the model. This will be decayed during training.
:param deconvolution: If set to True, will use transpose convolution(deconvolution) instead of up-sampling. This
increases the amount memory required during training.
:return: Untrained 3D UNet Model
"""
inputs = Input(input_shape)
current_layer = inputs
levels = list()
# add levels with max pooling
for layer_depth in range(depth):
layer1 = create_convolution_block(
input_layer=current_layer,
n_filters=n_base_filters * (2**layer_depth),
batch_normalization=batch_normalization)
layer2 = create_convolution_block(
input_layer=layer1,
n_filters=n_base_filters * (2**layer_depth) * 2,
batch_normalization=batch_normalization)
if layer_depth < depth - 1:
current_layer = MaxPooling3D(pool_size=pool_size)(layer2)
levels.append([layer1, layer2, current_layer])
else:
current_layer = layer2
levels.append([layer1, layer2])
# add levels with up-convolution or up-sampling
for layer_depth in range(depth - 2, -1, -1):
up_convolution = get_up_convolution(
pool_size=pool_size,
deconvolution=deconvolution,
n_filters=current_layer._keras_shape[1])(current_layer)
concat = concatenate([up_convolution, levels[layer_depth][1]], axis=1)
current_layer = create_convolution_block(
n_filters=levels[layer_depth][1]._keras_shape[1],
input_layer=concat,
batch_normalization=batch_normalization)
current_layer = create_convolution_block(
n_filters=levels[layer_depth][1]._keras_shape[1],
input_layer=current_layer,
batch_normalization=batch_normalization)
final_convolution = Conv3D(n_labels, (1, 1, 1))(current_layer)
act = Activation(activation_name)(final_convolution)
model = Model(inputs=inputs, outputs=act)
if not isinstance(metrics, list):
metrics = [metrics]
if include_label_wise_dice_coefficients and n_labels > 1:
label_wise_dice_metrics = [
get_label_dice_coefficient_function(index)
for index in range(n_labels)
]
if metrics:
metrics = metrics + label_wise_dice_metrics
else:
metrics = label_wise_dice_metrics
model.compile(optimizer=Adam(lr=initial_learning_rate),
loss=dice_coefficient_loss,
metrics=metrics)
return model
if Cosine_scheduler:
# 预热期
warmup_epoch = int((Freeze_epoch-Init_epoch)*0.2)
# 总共的步长
total_steps = int((Freeze_epoch-Init_epoch) * num_train / batch_size)
# 预热步长
warmup_steps = int(warmup_epoch * num_train / batch_size)
# 学习率
reduce_lr = WarmUpCosineDecayScheduler(learning_rate_base=learning_rate_base,
total_steps=total_steps,
warmup_learning_rate=1e-4,
warmup_steps=warmup_steps,
hold_base_rate_steps=num_train,
min_learn_rate=1e-6
)
model.compile(optimizer=Adam(), loss={'yolo_loss': lambda y_true, y_pred: y_pred})
else:
reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=2, verbose=1)
model.compile(optimizer=Adam(learning_rate_base), loss={'yolo_loss': lambda y_true, y_pred: y_pred})
print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size))
model.fit_generator(data_generator(lines[:num_train], batch_size, input_shape, anchors, num_classes, mosaic=mosaic),
steps_per_epoch=max(1, num_train//batch_size),
validation_data=data_generator(lines[num_train:], batch_size, input_shape, anchors, num_classes, mosaic=False),
validation_steps=max(1, num_val//batch_size),
epochs=Freeze_epoch,
initial_epoch=Init_epoch,
callbacks=[logging, checkpoint, reduce_lr, early_stopping])
model.save_weights(log_dir + 'trained_weights_stage_1.h5')
for i in range(freeze_layers): model_body.layers[i].trainable = True
x = Dense(256, activation='relu')(x)
predictions = Dense(num_classes, activation='softmax')(x)
model = Model(inputs=base_model.input, outputs=predictions)
base_model.trainable = True
set_trainable = False
for layer in base_model.layers:
if layer.name == 'conv2d_90':
set_trainable = True
if set_trainable:
layer.trainable = True
else:
layer.trainable = False
model.compile(optimizer=Adam(lr=0.00001),
loss='categorical_crossentropy',
metrics=['accuracy'])
model.summary()
# checkpoints
stem_id = "7_2_1_InceptionV3-unfrozen1-mask-imagenet-imgsize"
filepath = stem_id + str(
size) + ".best_{epoch:02d}-{val_accuracy:.2f}.hdf5"
checkpoint = ModelCheckpoint(filepath,
monitor='val_accuracy',
verbose=1,
save_best_only=True,
mode='max')
filepath = stem_id + str(size) + ".last_auto4.hdf5"
checkpoint_all = ModelCheckpoint(filepath,
total_weight_per_user = train.nnz / float(num_users)
train_csr, user_weights = train.tocsr(), []
for u in xrange(num_users):
#user_weights.append(total_weight_per_user / float(train_csr.getrow(u).nnz))
user_weights.append(1)
print("Load data done [%.1f s]. #user=%d, #item=%d, #train=%d, #test=%d"
%(time()-t1, num_users, num_items, train.nnz, len(testRatings)))
# Build model
model = get_model(num_users, num_items, mf_dim, layers, reg_layers, reg_mf, enable_dropout)
if learner.lower() == "adagrad":
model.compile(optimizer=Adagrad(lr=learning_rate), loss='binary_crossentropy')
elif learner.lower() == "rmsprop":
model.compile(optimizer=RMSprop(lr=learning_rate), loss='binary_crossentropy')
elif learner.lower() == "adam":
model.compile(optimizer=Adam(lr=learning_rate), loss='binary_crossentropy')
else:
model.compile(optimizer=SGD(lr=learning_rate), loss='binary_crossentropy')
# Load pretrain model
if mf_pretrain != '' and mlp_pretrain != '':
gmf_model = GMFlogistic.get_model(num_users,num_items,mf_dim)
gmf_model.load_weights(mf_pretrain)
mlp_model = MLPlogistic.get_model(num_users,num_items, layers, reg_layers)
mlp_model.load_weights(mlp_pretrain)
model = load_pretrain_model(model, gmf_model, mlp_model, len(layers))
print("Load pretrained GMF (%s) and MLP (%s) models done. " %(mf_pretrain, mlp_pretrain))
# Init performance
(hits, ndcgs) = evaluate_model(model, testRatings, testNegatives, topK, evaluation_threads)
hr, ndcg = np.array(hits).mean(), np.array(ndcgs).mean()
model.add(Dense(50, activation = 'elu', name = 'FC2'))
model.add(Dropout(0.2))
model.add(Dense(10, activation = 'elu', name = 'FC3'))
model.add(Dropout(0.2))
model.add(Dense(1, activation = 'elu', name = 'FC4'))
model.summary()
# checkpoints
checkpoint = ModelCheckpoint("./model_nvidia- {epoch:003d}.h5",
monitor = 'val_loss',
verbose = 1,
save_best_only = True,
mode = 'auto')
# compile
opt = Adam(lr = 0.001)
model.compile(optimizer = opt, loss = 'mse', metrics = [])
class LifecycleCallback(keras.callbacks.Callback):
def on_epoch_begin(self, epoch, logs = {}):
pass
def on_epoch_end(self, epoch, logs = {}):
global threshold
threshold = 1 / (epoch + 1)
def on_batch_begin(self, batch, logs = {}):
pass
def on_batch_end(self, batch, logs = {}):
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Runs a trained in a preprocessed dataset (npz)')
parser.add_argument('-input_model', help='path of the model')
parser.add_argument('-input_data', help='path of the input data')
parser.add_argument('-output_csv', help='path of the output csv')
parser.add_argument('--roi_statistics_csv', default = '', help=' (OPTIONAL) Annotate statistics')
parser.add_argument('--threshold', type = float, default = -1, help=' (OPTIONAL) Discard patches with less than that.')
parser.add_argument('--overwrite', action='store_true', help=' (OPTIONAL) Overwrite Default none.')
parser.add_argument('--convertToFloat', action='store_true', help=' (OPTIONAL) Transform the images to float. Dunno why, but some networks only work with one kind (Mingot ones with float, new ones with int16).')
args = parser.parse_args()
#Load the network
K.set_image_dim_ordering('th')
model = ResnetBuilder().build_resnet_50((3,40,40),1)
model.compile(optimizer=Adam(lr=1e-4), loss='mse', metrics=['mse'])
logging.info('Loading existing model %s...' % args.input_model)
model.load_weights(args.input_model)
#Create a dataframe for the ROIS
stats_roi_pd = pd.DataFrame()
#Get the patient files
if os.path.isdir(args.input_data):
patientFiles = map(lambda s: os.path.join(args.input_data, s) ,filter(lambda s: s.endswith('.npz'), os.listdir(args.input_data)))
else:
patientFiles = []
with open(args.input_data, 'r') as f:
for line in f:
patientFiles.append(line.strip())
def train_result(x_train, y_train, x_test, y_test, name, Results):
inputs = Input((224, 224, 1))
conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(inputs)
conv1 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(pool1)
conv2 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv2)
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
conv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(pool2)
conv3 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv3)
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
conv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(pool3)
conv4 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv4)
pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)
conv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(pool4)
conv5 = Conv2D(512, (3, 3), activation='relu', padding='same')(conv5)
up6 = concatenate([
Conv2DTranspose(256,
(2, 2), strides=(2, 2), padding='same')(conv5), conv4
],
axis=3)
conv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(up6)
conv6 = Conv2D(256, (3, 3), activation='relu', padding='same')(conv6)
up7 = concatenate([
Conv2DTranspose(128,
(2, 2), strides=(2, 2), padding='same')(conv6), conv3
],
axis=3)
conv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(up7)
conv7 = Conv2D(128, (3, 3), activation='relu', padding='same')(conv7)
up8 = concatenate([
Conv2DTranspose(64,
(2, 2), strides=(2, 2), padding='same')(conv7), conv2
],
axis=3)
conv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(up8)
conv8 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv8)
up9 = concatenate([
Conv2DTranspose(32,
(2, 2), strides=(2, 2), padding='same')(conv8), conv1
],
axis=3)
conv9 = Conv2D(32, (3, 3), activation='relu', padding='same')(up9)
conv9 = Conv2D(32, (3, 3), activation='relu', padding='same')(conv9)
conv10 = Conv2D(1, (1, 1), activation='sigmoid')(conv9)
model = Model(inputs=[inputs], outputs=[conv10])
model.compile(optimizer=Adam(lr=1e-5),
loss=jaccard_distance_loss,
metrics=[dice_coef])
history = model.fit(x_train,
y_train,
batch_size=8,
validation_data=(x_test, y_test),
shuffle=True,
epochs=140)
History.append(history)
y_pred = model.predict(x_test)
print(y_pred.shape)
score = dice_score(y_test, y_pred)
print("dice score: ", score)
Dice_Scores.append((name, score))
for i in range(y_pred.shape[0]):
y_pred[i, :, :, :][y_pred[i, :, :, :] > 0.5] = 1
y_pred[i, :, :, :][y_pred[i, :, :, :] <= 0.5] = 0
Results = save_result(y_pred, name, Results)
return Results
def create_model(seed, epochs, batch_size):
train_generator2 = train_datagen2.flow_from_directory(
'data/train',
target_size=img_size,
batch_size=batch_size,
class_mode='categorical',
seed=seed)
validation_generator2 = test_datagen.flow_from_directory(
'data/validation',
target_size=img_size,
batch_size=batch_size,
class_mode='categorical',
seed=seed)
reset_random_seeds(seed)
model = Sequential([
Conv2D(baseMapNum, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay),
input_shape=(32, 32, 3)),
Activation('relu'),
BatchNormalization(),
Conv2D(baseMapNum, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay)),
Activation('relu'),
BatchNormalization(),
MaxPool2D(pool_size=(2, 2)),
Dropout(0.2),
Conv2D(2 * baseMapNum, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay)),
Activation('relu'),
BatchNormalization(),
Conv2D(2 * baseMapNum, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay)),
Activation('relu'),
BatchNormalization(),
MaxPool2D(pool_size=(2, 2)),
Dropout(0.3),
Conv2D(4 * baseMapNum, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay)),
Activation('relu'),
BatchNormalization(),
Conv2D(4 * baseMapNum, (3, 3), padding='same', kernel_regularizer=regularizers.l2(weight_decay)),
Activation('relu'),
BatchNormalization(),
MaxPool2D(pool_size=(2, 2)),
Dropout(0.4),
Flatten(),
Dense(128, activation='relu'),
BatchNormalization(),
Dropout(0.4),
Dense(num_classes, activation='softmax')
])
lrr = ReduceLROnPlateau(
monitor='val_accuracy',
factor=.5,
patience=8,
min_lr=1e-4,
verbose=1)
opt_adam = Adam(learning_rate=0.002, beta_1=0.9, beta_2=0.999)
model.compile(loss='categorical_crossentropy',
optimizer=opt_adam,
metrics=['accuracy'])
history = model.fit(train_generator2, epochs=epochs, validation_data=validation_generator2, callbacks=[lrr])
loss, acc = model.evaluate(validation_generator2)
return model, history, loss, acc
#model.add(Dropout(0.6))
#model.add(Conv1D(activation="relu", padding="valid", strides=1, filters=640, kernel_size=3, kernel_initializer='glorot_uniform', kernel_regularizer=l2(0.001)))
#model.add(MaxPooling1D(pool_size=2))
#model.add(Dropout(0.5))
#model.add(Flatten())
#model.summary()
model.add(Dense(units=512, input_dim=2080, activation="relu", kernel_initializer='glorot_uniform'))
model.add(Dropout(0.5))
#model.add(Dense(units=512, input_dim=512, activation="relu", kernel_initializer='glorot_uniform',kernel_regularizer=l2(0.001)))
#model.add(Dropout(0.5))
model.add(Dense(units=180, activation="relu",kernel_initializer='glorot_uniform'))
model.add(Dropout(0.5))
model.add(Dense(units=70, activation="relu",kernel_initializer='glorot_uniform'))
model.add(Dense(units=1, activation="sigmoid"))
model.summary()
adam = Adam(lr=0.0001)
sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
model.compile(loss='binary_crossentropy', optimizer=adam, metrics=['accuracy'])
print('running at most 60 epochs')
checkpointer = ModelCheckpoint(filepath="HistoneMark_H3K27ac.hdf5", verbose=1, save_best_only=True)
earlystopper = EarlyStopping(monitor='val_loss', patience=10, verbose=1)
model.fit(X_train_H3K27ac, Y_train_H3K27ac, batch_size=128, epochs=50, shuffle=True, validation_data=( X_val_H3K27ac, Y_val_H3K27ac), callbacks=[checkpointer,earlystopper])
#model.fit(X_train_s, Y_train_s, batch_size=12, epochs=50, shuffle=True, validation_data=( X_val_s, Y_val_s), callbacks=[checkpointer,earlystopper])
y_pred = model.predict(X_test_H3K27ac)
#y_pred = model.predict(X_test_s)
#tresults = model.evaluate(X_test_s, Y_test_s)
np.savetxt('H3K27ac_true.csv', Y_test_H3K27ac, delimiter=",")
np.savetxt('H3K27ac_pred.csv', y_pred, delimiter=",")
tresults = model.evaluate(X_test_H3K27ac, Y_test_H3K27ac)
print(tresults)
model.summary()
# read config
sentence_words_num = int(getConfig('cnn', 'WORDS_DIM'))
word_vector_dim = int(getConfig('cnn', 'VEC_DIM'))
# end config
model = Sequential()
model.add(Conv1D(500, 3, activation='relu', input_shape=(300, 900)))
model.add(MaxPooling1D(3))
model.add(Dropout(1.0))
model.add(Flatten())
model.add(Dense(11, activation='softmax'))
# sgd = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
adam = Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy'])
# test
early_stopping = EarlyStopping(monitor='acc', patience=3, mode='max')
model.fit(x_train, y_train, batch_size=128, epochs=50, callbacks=[early_stopping])
score_test = model.evaluate(x_test, y_test, batch_size=64)
print 'loss=', score_test[0], ' acc=', score_test[1]
# dev
score_dev = model.evaluate(x_dev, y_dev, batch_size=64)
print 'dev loss=', score_dev[0], ' acc=', score_dev[1]
now_time = GetNowTime()
# logging
logging.basicConfig(level=logging.DEBUG,
MaxPooling2D(pool_size=2),
Flatten(),
Dense(1024, activation='relu'),
Dropout(0.2),
BatchNormalization(),
Dense(1024, activation='relu'),
Dropout(0.2),
BatchNormalization(),
Dense(256, activation='relu'),
Dropout(0.2),
BatchNormalization(),
Dense(NUM_CLASSES, activation='softmax')
])
cnn_model.compile(loss='sparse_categorical_crossentropy',
optimizer=Adam(lr=0.001),
metrics=['accuracy'])
# summary
cnn_model.summary()
# Fitting the model
cnn_model.fit(x_train,
y_train,
batch_size=BATCH_SIZE,
epochs=EPOCHS,
verbose=1,
validation_data=(x_validate, y_validate))
cnn_model.save('alexNet_model.h5')
GAMMA=1 # GAMMA of our cumulative reward function
STEPS_PER_EPISODE = 30 # No. of time-steps per episode
# configure and compile our agent by using built-in Keras optimizers and the metrics!
# allocate the memory by specifying the maximum no. of samples to store
memory = SequentialMemory(limit=300000, window_length=1)
# random process for exploration noise
random_process = OrnsteinUhlenbeckProcess(size=nb_actions, theta=theta, mu=0., dt=0.01, sigma=sigma)
# define the DDPG agent
agent = DDPGAgent(nb_actions=nb_actions, actor=actor, critic=critic, critic_action_input=action_input,
memory=memory, nb_steps_warmup_critic=100, nb_steps_warmup_actor=100,
random_process=random_process, gamma=GAMMA, target_model_update=1e-3)
# compile the model as follows
agent.compile(Adam(lr=.001, clipnorm=1.), metrics=['mse'])
callbacks = common_func.build_callbacks(ENV_NAME, log_filename_pre, filename_exp)
# ----------------------------------------------------------------------------------------------------------------------------------------
# Training phase
# fitting the agent. After training is done, save the final weights.
# 240000
# agent.fit(env, nb_steps=300000, visualize=False, callbacks=callbacks, verbose=1, nb_max_episode_steps=STEPS_PER_EPISODE, process_noise_std=process_noise_std)
# agent.save_weights(log_filename_pre+filename_exp+'/ddpg_{}_weights.h5f'.format(ENV_NAME), overwrite=True)
# common_func.save_process_noise(ENV_NAME, log_filename_pre, filename_exp, process_noise_std, theta)
# -----------------------------------------------------------------------------------------------------------------------------------------
# Testing phase
# std_dev_noise： actuator noise while testing
agent.load_weights(log_filename_pre+filename_exp+'/ddpg_{}_weights.h5f'.format(ENV_NAME))
params += _params
regularizers += _regularizers
constraints += _consts
updates += _updates
print("parameters:")
print(params)
print("regularizers:")
print(regularizers)
print("constrains:")
print(constraints)
print("updates:")
print(updates)
"""updates"""
optimizer = Adam()
_updates = optimizer.get_updates(params, constraints, train_loss)
updates += _updates
print("after Adam, updates:")
for update in updates:
print(update)
train_ins = [X_train, y, weights]
test_ins = [X_test, y, weights]
predict_ins = [X_test]
"""Get functions"""
print("complie: _train")
_train = K.function(train_ins, [train_loss], updates=updates)
print("complie: _train_with_acc")
def run_cifar10(batch_size,
nb_epoch,
depth,
nb_dense_block,
nb_filter,
growth_rate,
dropout_rate,
learning_rate,
weight_decay,
plot_architecture):
""" Run CIFAR10 experiments
:param batch_size: int -- batch size
:param nb_epoch: int -- number of training epochs
:param depth: int -- network depth
:param nb_dense_block: int -- number of dense blocks
:param nb_filter: int -- initial number of conv filter
:param growth_rate: int -- number of new filters added by conv layers
:param dropout_rate: float -- dropout rate
:param learning_rate: float -- learning rate
:param weight_decay: float -- weight decay
:param plot_architecture: bool -- whether to plot network architecture
"""
###################
# Data processing #
###################
# the data, shuffled and split between train and test sets
(X_train, y_train), (X_test, y_test) = cifar10.load_data()
nb_classes = len(np.unique(y_train))
img_dim = X_train.shape[1:]
if K.image_dim_ordering() == "th":
n_channels = X_train.shape[1]
else:
n_channels = X_train.shape[-1]
# convert class vectors to binary class matrices
Y_train = np_utils.to_categorical(y_train, nb_classes)
Y_test = np_utils.to_categorical(y_test, nb_classes)
X_train = X_train.astype('float32')
X_test = X_test.astype('float32')
# Normalisation
X = np.vstack((X_train, X_test))
# 2 cases depending on the image ordering
if K.image_dim_ordering() == "th":
for i in range(n_channels):
mean = np.mean(X[:, i, :, :])
std = np.std(X[:, i, :, :])
X_train[:, i, :, :] = (X_train[:, i, :, :] - mean) / std
X_test[:, i, :, :] = (X_test[:, i, :, :] - mean) / std
elif K.image_dim_ordering() == "tf":
for i in range(n_channels):
mean = np.mean(X[:, :, :, i])
std = np.std(X[:, :, :, i])
X_train[:, :, :, i] = (X_train[:, :, :, i] - mean) / std
X_test[:, :, :, i] = (X_test[:, :, :, i] - mean) / std
###################
# Construct model #
###################
model = densenet.DenseNet(nb_classes,
img_dim,
depth,
nb_dense_block,
growth_rate,
nb_filter,
dropout_rate=dropout_rate,
weight_decay=weight_decay)
# Model output
model.summary()
# Build optimizer
opt = Adam(lr=learning_rate, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
model.compile(loss='categorical_crossentropy',
optimizer=opt,
metrics=["accuracy"])
if plot_architecture:
from keras.utils.visualize_util import plot
plot(model, to_file='./figures/densenet_archi.png', show_shapes=True)
####################
# Network training #
####################
print("Training")
list_train_loss = []
list_test_loss = []
list_learning_rate = []
for e in range(nb_epoch):
if e == int(0.5 * nb_epoch):
K.set_value(model.optimizer.lr, np.float32(learning_rate / 10.))
if e == int(0.75 * nb_epoch):
K.set_value(model.optimizer.lr, np.float32(learning_rate / 100.))
split_size = batch_size
num_splits = X_train.shape[0] / split_size
arr_splits = np.array_split(np.arange(X_train.shape[0]), num_splits)
l_train_loss = []
start = time.time()
for batch_idx in arr_splits:
X_batch, Y_batch = X_train[batch_idx], Y_train[batch_idx]
train_logloss, train_acc = model.train_on_batch(X_batch, Y_batch)
l_train_loss.append([train_logloss, train_acc])
test_logloss, test_acc = model.evaluate(X_test,
Y_test,
verbose=0,
batch_size=64)
list_train_loss.append(np.mean(np.array(l_train_loss), 0).tolist())
list_test_loss.append([test_logloss, test_acc])
list_learning_rate.append(float(K.get_value(model.optimizer.lr)))
# to convert numpy array to json serializable
print('Epoch %s/%s, Time: %s' % (e + 1, nb_epoch, time.time() - start))
d_log = {}
d_log["batch_size"] = batch_size
d_log["nb_epoch"] = nb_epoch
d_log["optimizer"] = opt.get_config()
d_log["train_loss"] = list_train_loss
d_log["test_loss"] = list_test_loss
d_log["learning_rate"] = list_learning_rate
json_file = os.path.join('./log/experiment_log_cifar10.json')
with open(json_file, 'w') as fp:
json.dump(d_log, fp, indent=4, sort_keys=True)
def __init__(self):
#self.path = "/volumes/data/dataset/gan/MNIST/wgan-gp/wgan-gp_generated_images/"
self.path = "images/"
#mnistデータ用の入力データサイズ
self.img_rows = 28
self.img_cols = 28
self.channels = 1
self.img_shape = (self.img_rows, self.img_cols, self.channels)
# 潜在変数の次元数
self.z_dim = 5
self.n_critic = 5
# 画像保存の際の列、行数
self.row = 5
self.col = 5
self.row2 = 1 # 連続潜在変数用
self.col2 = 10# 連続潜在変数用
# 画像生成用の固定された入力潜在変数
self.noise_fix1 = np.random.normal(0, 1, (self.row * self.col, self.z_dim))
# 連続的に潜在変数を変化させる際の開始、終了変数
self.noise_fix2 = np.random.normal(0, 1, (1, self.z_dim))
self.noise_fix3 = np.random.normal(0, 1, (1, self.z_dim))
# 横軸がiteration数のプロット保存用np.ndarray
self.g_loss_array = np.array([])
self.d_loss_array = np.array([])
self.d_accuracy_array = np.array([])
self.d_predict_true_num_array = np.array([])
self.c_predict_class_list = []
#discriminator_optimizer = Adam(lr=1e-5, beta_1=0.1)
combined_optimizer = Adam(lr=1e-4, beta_1=0.5, beta_2=0.9)
# discriminatorモデル
self.discriminator = self.build_discriminator()
# Generatorモデル
self.generator = self.build_generator()
# combinedモデルの学習時はdiscriminatorの学習をFalseにする
for layer in self.discriminator.layers:
layer.trainable = False
self.discriminator.trainable = False
self.netG_model, self.netG_train = self.build_combined()
for layer in self.discriminator.layers:
layer.trainable = True
for layer in self.generator.layers:
layer.trainable = False
self.discriminator.trainable = True
self.generator.trainable = False
# Classifierモデル
self.classifier = self.build_classifier()
self.netD_train = self.build_discriminator_with_own_loss()
print(type(x_train))
#
x_train /= 255
x_test /= 255
#print(x_train[0])
print(y_train[0])
#把整形int标签转换成one-hot编码的数组标签,以方便计算loss
y_train = keras.utils.to_categorical(y_train, n_classes)
y_test = keras.utils.to_categorical(y_test, n_classes)
print(y_train[0])
#搭建模型
model=Sequential()
model.add(LSTM(n_hidden,batch_input_shape=(None,n_step,n_input),unroll=True))
model.add(Dense(n_classes))#这个参数应该与输出维度相同了
model.add(Activation('softmax'))
adam=Adam(lr=learning_rate)
model.summary()
model.compile(optimizer=adam,
loss='categorical_crossentropy',
metrics=['accuracy'])
model.fit(x_train,y_train,batch_size=batch_size,epochs=training_iters,verbose=1,validation_data=(x_test,y_test))
score=model.evaluate(x_test,y_test,verbose=0)
print('LSTM test score:',score[0])
print('LSTM test accuracy:',score[1])
def get_mitosegnet(self, wmap, lr):
inputs = Input(shape=(self.img_rows, self.img_cols, 1))
print(inputs.get_shape(), type(inputs))
# core mitosegnet (modified u-net) architecture
# batchnorm architecture (batchnorm before activation)
######################################################################
conv1 = Conv2D(64, 3, padding='same',
kernel_initializer=gauss())(inputs)
print("conv1 shape:", conv1.shape)
batch1 = BatchNormalization()(conv1)
act1 = Activation("relu")(batch1)
conv1 = Conv2D(64,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 64))))(
act1) # conv1
print("conv1 shape:", conv1.shape)
batch1 = BatchNormalization()(conv1)
act1 = Activation("relu")(batch1)
pool1 = MaxPooling2D(pool_size=(2, 2))(act1)
print("pool1 shape:", pool1.shape)
########
########
conv2 = Conv2D(128,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 64))))(pool1)
print("conv2 shape:", conv2.shape)
batch2 = BatchNormalization()(conv2)
act2 = Activation("relu")(batch2)
conv2 = Conv2D(128,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 128))))(
act2) # conv2
print("conv2 shape:", conv2.shape)
batch2 = BatchNormalization()(conv2)
act2 = Activation("relu")(batch2)
pool2 = MaxPooling2D(pool_size=(2, 2))(act2)
print("pool2 shape:", pool2.shape)
########
########
conv3 = Conv2D(256,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 128))))(pool2)
print("conv3 shape:", conv3.shape)
batch3 = BatchNormalization()(conv3)
act3 = Activation("relu")(batch3)
conv3 = Conv2D(256,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 256))))(
act3) # conv3
print("conv3 shape:", conv3.shape)
batch3 = BatchNormalization()(conv3)
act3 = Activation("relu")(batch3)
pool3 = MaxPooling2D(pool_size=(2, 2))(act3)
print("pool3 shape:", pool3.shape)
########
########
conv4 = Conv2D(512,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 256))))(pool3)
batch4 = BatchNormalization()(conv4)
act4 = Activation("relu")(batch4)
conv4 = Conv2D(512,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 512))))(
act4) # conv4
batch4 = BatchNormalization()(conv4)
act4 = Activation("relu")(batch4)
pool4 = MaxPooling2D(pool_size=(2, 2))(act4)
########
########
conv5 = Conv2D(1024,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 512))))(pool4)
batch5 = BatchNormalization()(conv5)
act5 = Activation("relu")(batch5)
conv5 = Conv2D(1024,
3,
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 1024))))(
act5) # conv5
batch5 = BatchNormalization()(conv5)
act5 = Activation("relu")(batch5)
########
up6 = Conv2D(512,
2,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 1024))))(
UpSampling2D(size=(2, 2))(act5))
merge6 = concatenate([conv4, up6], axis=3)
conv6 = Conv2D(
512,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 512))))(merge6)
conv6 = Conv2D(512,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 512))))(conv6)
up7 = Conv2D(256,
2,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 512))))(
UpSampling2D(size=(2, 2))(conv6))
merge7 = concatenate([conv3, up7], axis=3)
conv7 = Conv2D(
256,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 256))))(merge7)
conv7 = Conv2D(256,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 256))))(conv7)
up8 = Conv2D(128,
2,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 256))))(
UpSampling2D(size=(2, 2))(conv7))
merge8 = concatenate([conv2, up8], axis=3)
conv8 = Conv2D(
128,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 128))))(merge8)
conv8 = Conv2D(128,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 128))))(conv8)
up9 = Conv2D(64,
2,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 / (9 * 128))))(
UpSampling2D(size=(2, 2))(conv8))
merge9 = concatenate([conv1, up9], axis=3)
conv9 = Conv2D(64,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 64))))(merge9)
conv9 = Conv2D(64,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 64))))(conv9)
conv9 = Conv2D(2,
3,
activation='relu',
padding='same',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 64))))(conv9)
######################################################################
conv10 = Conv2D(1,
1,
activation='sigmoid',
kernel_initializer=gauss(stddev=sqrt(2 /
(9 * 2))))(conv9)
if wmap == False:
input = inputs
loss = self.pixelwise_crossentropy()
else:
weights = Input(shape=(self.img_rows, self.img_cols, 1))
input = [inputs, weights]
loss = self.weighted_pixelwise_crossentropy(input[1])
model = Model(inputs=input, outputs=conv10)
model.compile(optimizer=Adam(lr=lr),
loss=loss,
metrics=['accuracy', self.dice_coefficient])
return model
# Using the Glove embedding:
i = 0
for word in vocabulary:
embedding_vector = embeddings_index.get(word[0])
if embedding_vector is not None:
# words not found in embedding index will be all-zeros.
embedding_matrix[i] = embedding_vector
i += 1
# *******************************************************************
# Keras model of the chatbot:
# *******************************************************************
ad = Adam(lr=0.00005)
input_context = Input(shape=(maxlen_input,), dtype='int32')#, name='input_context')
input_answer = Input(shape=(maxlen_input,), dtype='int32')#, name='input_answer')
LSTM_encoder = LSTM(sentence_embedding_size, kernel_initializer= 'lecun_uniform')
LSTM_decoder = LSTM(sentence_embedding_size, kernel_initializer= 'lecun_uniform')
if os.path.isfile(weights_file):
Shared_Embedding = Embedding(output_dim=word_embedding_size, input_dim=dictionary_size, input_length=maxlen_input)
else:
Shared_Embedding = Embedding(output_dim=word_embedding_size, input_dim=dictionary_size, weights=[embedding_matrix], input_length=maxlen_input)
word_embedding_context = Shared_Embedding(input_context)
context_embedding = LSTM_encoder(word_embedding_context)
# LSTM_encoder_topic = LSTM(topic_embedding_size, kernel_initializer='lecun_uniform')
LSTM_encoder_topic = Dense(topic_embedding_size, activation="relu")
topic_embedding = LSTM_encoder_topic(context_embedding)
if X_train_seq_res_std[i, j] == 0:
X_train_seq_res[:,i, j] = 0
else:
X_train_seq_res[:, i, j] = (X_train_seq_res[:, i, j] - X_train_seq_res_mean[i, j])/X_train_seq_res_std[i, j]
#########################
# MODEL SETUP #
#########################
# Set training hyper-parameters.
epochs = 22
batch_size = 64
learn_rate = 0.001
drop_prob = 0.75
optimiser = Adam(lr=learn_rate)
branch1 = Sequential()
branch1.add(BatchNormalization(input_shape=(num_seq_steps, num_seq_features)))
branch1.add(Conv1D(50, 5, kernel_initializer='he_uniform', activation='relu',
input_shape=(num_seq_steps, num_seq_features)))
branch1.add(MaxPooling1D(2))
branch1.add(Dropout(drop_prob))
branch1.add(BatchNormalization())
branch1.add(Conv1D(50, 11, kernel_initializer='he_uniform', activation='relu'))
branch1.add(MaxPooling1D(2))
branch1.add(Dropout(drop_prob))
branch1.add(BatchNormalization())
