def train(self, dataset, batch_size=20, epochs=10, data_up=True):
# 生成SGD优化器进行训练
sgd = SGD(lr=0.01, decay=1e-6,momentum=0.9, nesterov=True)
# 完成实际的模型配置工作
self.model.compile(loss='categorical_crossentropy',optimizer=sgd,metrics=['accuracy'])
# 数据提升
if not data_up:
self.model.fit(dataset.train_images,
dataset.train_labels,
batch_size=batch_size,
epochs=epochs,
validation_data=(dataset.valid_images, dataset.valid_labels),
shuffle=True)
# 使用实时数据提升
else:
# 定义数据生成器
datagen = ImageDataGenerator(
rotation_range=20, # 数据提升时图片随机转动的角度
width_shift_range=0.2, # 数据提升时图片水平偏移的幅度，百分比
height_shift_range=0.2, # 垂直平移幅度
horizontal_flip=True, # 随机水平翻转
vertical_flip=False) # 随机垂直翻转
# 计算整个训练样本集的数量
datagen.fit(dataset.train_images)
# 利用生成器开始训练模型
self.model.fit_generator(datagen.flow(dataset.train_images, dataset.train_labels,
batch_size=batch_size),
samples_per_epoch=dataset.train_images.shape[0],
epochs=epochs,
validation_data=(dataset.valid_images, dataset.valid_labels))
# 构造TensorBoard
tbCallBack = tb(log_dir="/Users/heyiheng/Desktop/biyelunwen/LastDemo/logs",
histogram_freq=1,
batch_size=batch_size,
write_grads=True)
history = self.model.fit(dataset.train_images, dataset.train_labels,
batch_size=batch_size,
epochs=epochs,
shuffle=True,
verbose=2,
validation_split=0.2,
callbacks=[tbCallBack])
return model, history
sequence_data = paddingsequence(tokenizer,MAX_SEQUENCE_LENGTH,strings)
#Train and Validation Creation
x_train,y_train,x_val,y_val = train_validation_create(sequence_data,labels,VALIDATION_SPLIT)
#####################Model Fitting##########################################
save_location = wd + "/Predictions/"
#Use pre-trained model to generate embedding
embedding_matrix = embeddingenerator(EMBEDDING_FILE,EMBEDDING_DIM,MAX_NB_WORDS,word_index)
#Initiating Tensorboard
tbCallBack = tb(log_dir= wd + "/Log/" + MODEL_NAME,histogram_freq=0, batch_size=32, write_graph=True, write_grads= True, write_images=True)
model = kerasmodelbuilder(EMBEDDING_DIM,MAX_SEQUENCE_LENGTH, \
word_index,MAX_NB_WORDS,embedding_matrix)
model.fit(x_train, y_train,validation_data=(x_val, y_val), \
epochs=epochs, \
batch_size=batch_size, \
callbacks = [tbCallBack]
)
y_val_proba = model.predict(x_val)
fpr, tpr, _ = metrics.roc_curve(y_val, y_val_proba)
auc = metrics.roc_auc_score(y_val, y_val_proba)
plt.plot(fpr,tpr,label="validation data, auc="+str(auc))
from model import VggRes64
import pandas as pd
from keras.callbacks import TensorBoard as tb, ModelCheckpoint
from keras.models import load_model
if __name__ == "__main__":
model_name = "discriminator_2x2_tiny_label_25.h5"
source_dir = "E:/Data/mtcnn_64/"
df = pd.read_csv("./source/mtcnn_64_drop_col.txt", sep=" ")
model = VggRes64.VggRes64()
model.set_model(load_model("./save/{}".format(model_name)))
tesnorboard = tb(log_dir='./logs/{}'.format(model_name), histogram_freq=0)
checkpoint = ModelCheckpoint('./save/{}'.format(model_name),
monitor='val_loss',
verbose=1,
save_best_only=True,
period=10)
model.set_tensorboard(tesnorboard)
model.set_checkpoint(checkpoint)
train_set = df[:-10000]
validation_set = df[-10000:]
generator_train = model.create_image_generator(train_set, "filename",
source_dir)
generator_validation = model.create_image_generator(
validation_set, "filename", source_dir)
model.fit_generator(generator_train, generator_validation, epochs=500)
# load YAML and create model
#with open(DATADIR+trained_model, 'r') as yaml_file:
# loaded_model_yaml = yaml_file.read()
# loaded_model = model_from_yaml(loaded_model_yaml)
# load weights into new model
#loaded_model.load_weights(DATADIR+trained_weights)
model = load_model(DATADIR + compiled_model)
print("Loaded model from disk")
from keras.callbacks import TensorBoard as tb
from datetime import datetime
t = datetime.now()
tensorboard = tb(
log_dir='tensorboard_logs/{:%Y-%m-%d-%H-%M}'.format(t)) # Dial
with h5py.File(DATADIR + training_file, "r") as f1:
with h5py.File(DATADIR + label_file, "r") as f2:
x_train = f1[
'training_data'] # Note that Keras is special in being able to read the HDF5 _object_
y_train = f2['training_labels']
model.fit(
x_train,
y_train,
batch_size=batch_size,
epochs=epochs,
shuffle='batch', # Required for using HDF5
callbacks=[tensorboard])
def crearConfigTensorBoard(pathToLogs):
return tb(log_dir=pathToLogs,
histogram_freq=0,
write_graph=True,
write_images=False)