def create_and_fit_model():
input_shape = (CONFIG.IMAGE_TARGET_HEIGHT, CONFIG.IMAGE_TARGET_WIDTH, 3)
model = create_cnn(input_shape, dropout=True)
model.summary()
logger.info(len(model.trainable_weights))
# Add checkpoint callback.
#best_model_path = os.path.join('validation','best_model.h5')
best_model_path = str(DATA_DIR / f'outputs/{MODEL_CKPT_FILENAME}')
checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
filepath=best_model_path,
monitor="val_loss",
save_best_only=True,
verbose=1)
dataset_batches = dataset_training.batch(CONFIG.BATCH_SIZE)
training_callbacks = [
AzureLogCallback(run),
create_tensorboard_callback(),
checkpoint_callback,
]
if getattr(CONFIG, 'USE_WANDB', False):
setup_wandb()
wandb.init(project="ml-project", entity="cgm-team")
wandb.config.update(CONFIG)
training_callbacks.append(
WandbCallback(log_weights=True,
log_gradients=True,
training_data=dataset_batches))
optimizer = tf.keras.optimizers.RMSprop(learning_rate=CONFIG.LEARNING_RATE)
# Compile the model.
model.compile(optimizer=optimizer,
loss="binary_crossentropy",
metrics=["accuracy"])
# Train the model.
model.fit(dataset_training.batch(CONFIG.BATCH_SIZE),
validation_data=dataset_validation.batch(CONFIG.BATCH_SIZE),
epochs=CONFIG.EPOCHS,
callbacks=training_callbacks,
verbose=2)
# function use to tune the top convolution layer
set_trainable_below_layers('block14_sepconv1', model)
model.fit(dataset_training.batch(CONFIG.BATCH_SIZE),
validation_data=dataset_validation.batch(CONFIG.BATCH_SIZE),
epochs=CONFIG.TUNE_EPOCHS,
callbacks=training_callbacks,
verbose=2)
def train_network(file = 'input/cnews.train.txt'):
"""训练部分
"""
acc_value = 0.0 # 验证集正确率
iterations = 0 # 迭代次数
loss, optimizer, acc = model.create_cnn()
# 用来保存进度
saver = tf.train.Saver()
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
# 若有上次的进度，则加载
if os.path.isfile(SAVER_DIC+'.index'):
saver.restore(sess, SAVER_DIC)
for i in range(TOTAL_ITERATIONS):
view_bar('训练进度：', i, TOTAL_ITERATIONS)
x_data, y_data = model.get_some_date(file, model.BATCH_SIZE) # 取批次数据
feed_dict = {model.x: x_data, model.y_: y_data, model.dropout_keep_prob: 0.5}
_, loss_test_value, acc_test_value = sess.run([optimizer, loss, acc], feed_dict=feed_dict)
# 输出 train的 loss 和正确率
with open(TRAIN_LOG, 'a') as f:
f.write(str(i) + ',' + str(loss_test_value) + ',' + str(acc_test_value) + '\n')
# 在验证集中测试
if i%VALID_ITERATION == 0:
model.USE_L2 = False # 关闭L2正则化
x_valid_data, y_valid_data = model.get_some_date(VALID_DIC, model.BATCH_SIZE)
feed_dict = {model.x: x_valid_data, model.y_: y_valid_data, model.dropout_keep_prob: 1}
_, loss_valid_value, acc_valid_value = sess.run([optimizer, loss, acc], feed_dict=feed_dict)
if acc_valid_value > acc_value:
acc_value = acc_valid_value
saver.save(sess, SAVER_DIC)
iterations = 0
# 输出 valid的 loss 和正确率
with open(VALID_LOG, 'a') as f:
f.write(str(i) + ',' + str(loss_valid_value) + ',' + str(acc_valid_value) + '\n')
else:
iterations += 1
# 正确率久不提升，则停止
if iterations >= DELAY_NUM:
break
model.USE_L2 = True # 启用L2正则化
def test(file='input/cnews.test.txt'):
"""测试部分
"""
loss, optimizer, acc = model.create_cnn()
# 用来读取参数
saver = tf.train.Saver()
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
# 若进度存在，则读取学习到的参数
if os.path.isfile(SAVER_DIC + '.index'):
saver.restore(sess, SAVER_DIC)
for i in range(ITERATIONS):
view_bar('测试进度：', i, ITERATIONS)
x_test_data, y_test_data = model.get_some_date(file, model.BATCH_SIZE)
feed_dict = {model.x: x_test_data, model.y_: y_test_data, model.dropout_keep_prob: 1}
loss_test_value, acc_test_value = sess.run([loss, acc], feed_dict=feed_dict)
# 输出 test loss 和正确率
with open(TEST_LOG, 'a') as f:
f.write(str(i) + ',' + str(loss_test_value) + ',' + str(acc_test_value) + '\n')
from keras.optimizers import Adam, SGD
import numpy as np
import random
import os
import model as model_arch
from keras.callbacks import ReduceLROnPlateau, TensorBoard, ModelCheckpoint, EarlyStopping
from ImageBatchGenerator import ImageBatchGenerator
from config import EPOCHS, INIT_LR, BS, IMG_HEIGTH_WIDTH, NUM_MATCH_NOT_MATCH_PAIR_CNT, TRAIN_IMAGE_PATH, TEST_IMAGE_PATH
if __name__ == '__main__':
random.seed(42)
print("[INFO] compiling model...")
model = model_arch.create_cnn(IMG_HEIGTH_WIDTH, IMG_HEIGTH_WIDTH, 3)
# sgd = SGD(lr=0.01, clipvalue=0.5)
opt = Adam(lr=INIT_LR, decay=INIT_LR / EPOCHS)
model.compile(loss="binary_crossentropy",
optimizer=opt,
metrics=["accuracy"])
# train the network
print("[INFO] training network...")
reduce_lr_on_plateau = ReduceLROnPlateau(monitor='val_acc',
factor=0.2,
patience=2,
verbose=0,
mode='auto',
min_delta=0.0001,
cooldown=0,
# Create dataset for activation
paths = paths_activate
dataset = tf.data.Dataset.from_tensor_slices(paths)
dataset_norm = dataset.map(
lambda path: tf_load_pickle(path, CONFIG.NORMALIZATION_VALUE))
dataset_norm = dataset_norm.cache()
dataset_norm = dataset_norm.prefetch(tf.data.experimental.AUTOTUNE)
dataset_activation = dataset_norm
del dataset_norm
# Note: Now the datasets are prepared.
# Create the model.
input_shape = (CONFIG.IMAGE_TARGET_HEIGHT, CONFIG.IMAGE_TARGET_WIDTH, 1)
model = create_cnn(input_shape, dropout=True)
model.summary()
best_model_path = str(DATA_DIR / f'outputs/{MODEL_CKPT_FILENAME}')
checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
filepath=best_model_path,
monitor="val_loss",
save_best_only=True,
verbose=1)
training_callbacks = [
AzureLogCallback(run),
create_tensorboard_callback(),
checkpoint_callback,
]
optimizer = tf.keras.optimizers.Nadam(learning_rate=CONFIG.LEARNING_RATE)
labels = lb.fit_transform(labels)
# 0 is match and 1 becomes not_match
# partition the data into training and testing splits using 80% of
# the data for training and the remaining 80% for testing
print(len(data_list1))
(trainX, testX, trainXX, testXX, trainY, testY) = train_test_split(data_list1,data_list2,labels, test_size=0.2, random_state=42)
# construct the image generator for data augmentation
aug = ImageDataGenerator(rotation_range=30, width_shift_range=0.1,
height_shift_range=0.1, shear_range=0.2, zoom_range=0.2,
horizontal_flip=True, fill_mode="nearest")
# initialize the model
print("[INFO] compiling model...")
cnn1 = model.create_cnn(56, 56, 3)
cnn2 = model.create_cnn(56, 56, 3)
# combining tensor output of the two cnn models
combinedInput = concatenate([cnn1.output, cnn2.output])
x = Dense(16, activation="relu")(combinedInput)
# binary predictor
y = Dense(1, activation="sigmoid")(x)
model = Model(inputs=[cnn1.input, cnn2.input], outputs=y)
opt = Adam(lr=INIT_LR, decay=INIT_LR / EPOCHS)
model.compile(loss="binary_crossentropy", optimizer=opt, metrics=["accuracy"])
def prepare_nn(self):
nn_model = create_cnn(self.input_shape, self.n_classes, self.loss,
self.optimizer, self.summary)
return nn_model
def gen_flow_for_two_inputs(X1, X2, y):
genX1 = image_generator.flow(X1, y, batch_size=BS, seed=666)
genX2 = image_generator.flow(X2, y, batch_size=BS, seed=666)
while True:
X1i = genX1.next()
X2i = genX2.next()
#Assert arrays are equal - this was for peace of mind, but slows down training
#np.testing.assert_array_equal(X1i[0],X2i[0])
yield [X1i[0], X2i[0]], X1i[1]
gen_flow = gen_flow_for_two_inputs(trainX, trainXX, trainY)
# initialize the model
print("[INFO] compiling model...")
cnn1 = model.create_cnn(img_heigth_width, img_heigth_width, 3)
cnn2 = model.create_cnn(img_heigth_width, img_heigth_width, 3)
# combining tensor output of the two cnn models
combinedInput = concatenate([cnn1.output, cnn2.output])
x = Dense(16, activation="relu")(combinedInput)
# binary predictor
y = Dense(1, activation="sigmoid")(x)
model = Model(inputs=[cnn1.input, cnn2.input], outputs=y)
sgd = SGD(lr=0.01, clipvalue=0.5)
opt = Adam(lr=INIT_LR, decay=INIT_LR / EPOCHS)
model.compile(loss="binary_crossentropy", optimizer=opt, metrics=["accuracy"])