def CreateCustomizedResNetModel(self, input_shape, classes):
X_input = Input(name='the_input', shape=input_shape)
level_h1 = block(32)(X_input)
level_m1 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(level_h1) # 池化层
level_h2 = block(64)(level_m1)
level_m2 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(level_h2) # 池化层
level_h3 = block(128)(level_m2)
level_m3 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(level_h3) # 池化层
# level_h4 = block(64)(level_m3)
# level_m4 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(level_h4) # 池化层
flayer = Flatten()(level_m3)
fc = Dense(classes, use_bias=True,
kernel_initializer='he_normal')(flayer) # 全连接层
y_pred = Activation('softmax', name='Activation0')(fc)
model = Model(inputs=X_input, outputs=y_pred)
optimizer = optimizers.Adadelta()
# model.compile(optimizer=optimizer, loss='binary_crossentropy')# [focal_loss])
model.compile(optimizer=optimizer,
loss=[focal_loss(alpha=0.25, gamma=2)])
print('Customized resnet model estabished.')
self.modelname = 'residual'
return model
def regularCNNForest(self, input_shape, classes, layer_counts):
X_input = Input(name='the_input', shape=input_shape)
layer_h1 = Conv2D(32, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=l2(0.0005))(X_input)
layer_h2 = Conv2D(32, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=l2(0.0005))(layer_h1)
layer_p2 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(layer_h2)
layer_h3 = Conv2D(64, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=l2(0.0005))(layer_p2)
layer_h4 = Conv2D(64, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=l2(0.0005))(layer_h3)
layer_p4 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(layer_h4)
layer_h5 = Conv2D(128, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=l2(0.0005))(layer_p4)
layer_h6 = Conv2D(128, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=l2(0.0005))(layer_h5)
layer_p6 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(layer_h6)
layer_h7 = Conv2D(256, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=None)(layer_p6)
layer_h8 = Conv2D(256, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=None)(layer_h7)
layer_p8 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(layer_h8)
layer_h9 = Conv2D(512, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=None)(layer_p8)
layer_h10 = Conv2D(512, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=None)(layer_h9)
# layer_p10 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(layer_h10)
layers = [layer_p2,layer_p4,layer_p6,layer_p8,layer_h10]
if layer_counts in (2,4,6,8,10):
output = layers[layer_counts//2-1]
else:
print('[ERROR]unresolved parameters: layer_counts.')
assert(0)
flayer = GlobalAveragePooling2D()(output)
fc2 = Dense(classes, use_bias=True, kernel_initializer='he_normal')(flayer) # 全连接层
y_pred = Activation('softmax', name='Activation0')(fc2)
model = Model(inputs=X_input, outputs=y_pred)
optimizer = optimizers.Adadelta()
model.compile(optimizer=optimizer, loss=[focal_loss(alpha=0.25, gamma=2)])
print('Regular cnn model with {} layers estabished.'.format(layer_counts))
self.modelname = 'layerInv'
return model
def CreateResNetModel(self, input_shape, classes):
X_input = Input(name='the_input', shape=input_shape)
x = Conv2D(kernel_size=3,
filters=16,
strides=1,
padding='same',
kernel_regularizer=regularizers.l2(0.01))(X_input)
x = BatchNormalization()(x)
x = Activation(relu)(x)
x = block(16)(x)
x = block(16)(x)
x = BatchNormalization()(x)
x = Activation(relu)(x)
# 28x28x48 -> 1x48
x = GlobalAveragePooling2D()(x)
# dropout for more robust learning
x = Dropout(0.2)(x)
last = Dense(units=classes,
activation='softmax',
kernel_regularizer=regularizers.l2(0.01))(x)
model = Model(inputs=X_input, outputs=last)
optimizer = optimizers.Adadelta()
model.compile(optimizer=optimizer,
loss=[focal_loss(alpha=0.25, gamma=2)])
print('Original resnet neural network established. ')
return model
def CreateCustomizedInceptionModel(self, input_shape, classes):
X_input = Input(name='the_input', shape=input_shape)
layer_h1 = Conv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.01))(X_input) # 卷积层
layer_h2 = Conv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.01))(layer_h1) # 卷积层
layer_h3 = MaxPooling2D(pool_size=3, strides=None,
padding="valid")(layer_h2) # 池化层
# inception_3a = inception_model(input=layer_h3, filters_1x1=32, filters_3x3_reduce=48, filters_3x3=64,filters_5x5_reduce=8, filters_5x5=16, filters_pool_proj=16)
# inception_3b = inception_model(input=inception_3a, filters_1x1=85, filters_3x3_reduce=85, filters_3x3=128,filters_5x5_reduce=21, filters_5x5=64, filters_pool_proj=42)
inception_3a = inception_model(input=layer_h3,
filters_1x1=64,
filters_3x3_reduce=96,
filters_3x3=128,
filters_5x5_reduce=16,
filters_5x5=32,
filters_pool_proj=32)
inception_3b = inception_model(input=inception_3a,
filters_1x1=128,
filters_3x3_reduce=128,
filters_3x3=192,
filters_5x5_reduce=32,
filters_5x5=96,
filters_pool_proj=64)
layer_h3 = MaxPooling2D(pool_size=3, strides=None,
padding="valid")(inception_3b) # 池化层
transmute = core.Flatten()(layer_h3)
linear = Dense(units=classes,
activation='softmax',
kernel_regularizer=l2(0.01))(transmute)
last = linear
model = Model(inputs=X_input, outputs=last)
optimizer = optimizers.Adadelta()
model.compile(optimizer=optimizer,
loss=[focal_loss(alpha=0.25, gamma=2)])
# model.compile(optimizer=optimizer,loss='binary_crossentropy')
print('Customized inception neural network established. ')
self.modelname = 'inception'
return model
def bnRegularCNN(self, input_shape, classes):
X_input = Input(name='the_input', shape=input_shape)
level_h1 = ReguBlock(32)(X_input)
level_m1 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(level_h1) # 池化层
level_h2 = ReguBlock(64)(level_m1)
level_m2 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(level_h2) # 池化层
level_h3 = ReguBlock(128)(level_m2)
level_m3 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(level_h3) # 池化层
flayer = GlobalAveragePooling2D()(level_m3)
fc = Dense(classes, use_bias=True, kernel_initializer='he_normal')(flayer) # 全连接层
y_pred = Activation('softmax', name='Activation0')(fc)
model = Model(inputs=X_input, outputs=y_pred)
optimizer = optimizers.Adadelta()
model.compile(optimizer=optimizer, loss=[focal_loss(alpha=0.25, gamma=2)])
print('Regular toy cnn model within BN layers estabished.')
self.modelname = 'layerInv'
return model
def TrainModel(self,
datapath,
epoch=2,
batch_size=32,
load_weights=False,
filename='model_set/speech_model25'):
assert (batch_size % CLASS_NUM == 0)
data = DataSpeech(datapath, 'train')
num_data = sum(data.DataNum) # 获取数据的数量
os.system('pkill tensorboard')
os.system('rm -rf ./checkpoints/files_summary/* ')
train_writter = tf.summary.FileWriter(
os.path.join(os.getcwd(), 'checkpoints', 'files_summary'))
os.system(
'tensorboard --logdir=/home/zhaok14/example/PycharmProjects/setsail/individual_spp/checkpoints/files_summary/ &'
)
print('\n')
print(90 * '*')
print(90 * '*')
iterations_per_epoch = min(
data.DataNum) // (batch_size // CLASS_NUM) + 1
# iterations_per_epoch = 2
print('trainer info:')
print('training data size: %d' % num_data)
print('increased epoches: ', epoch)
print('minibatch size: %d' % batch_size)
print('iterations per epoch: %d' % iterations_per_epoch)
with k.get_session() as sess:
train_writter.add_graph(sess.graph)
if load_weights == True:
try:
# modelpath = os.path.join(os.getcwd(), 'network&&weights', 'spectrogram', 'inception','spec_inception.h5')
self.model = load_model(modelpath,
custom_objects={
'focal_loss': focal_loss,
'focal_loss_fixed':
focal_loss()
})
print('Successfully loading the model.')
except:
print('Loading weights failed. Train from scratch.')
sess.run(tf.global_variables_initializer())
best_score = 0
for i in range(0, epoch):
iteration = 0
yielddatas = data.data_genetator(batch_size, epoch)
pbar = tqdm(yielddatas)
for input, labels in pbar:
loss = self.model.train_on_batch(input[0], labels)
train_summary = tf.Summary()
train_summary.value.add(tag='loss', simple_value=loss)
train_writter.add_summary(
train_summary, iteration + i * iterations_per_epoch)
pr = 'epoch:%d/%d,iteration: %d/%d ,loss: %s' % (
epoch, i, iterations_per_epoch, iteration, loss)
pbar.set_description(pr)
if iteration == iterations_per_epoch:
break
else:
iteration += 1
pbar.close()
if i % 1 == 0:
self.TestModel(sess=sess,
datapath=self.datapath,
str_dataset='train',
data_count=1000,
out_report=False,
writer=train_writter,
step=i)
metrics = self.TestModel(sess=sess,
datapath=self.datapath,
str_dataset='eval',
data_count=-1,
out_report=False,
writer=train_writter,
step=i)
if (metrics['score'] >= best_score and i > 0):
self.metrics = metrics
self.metrics['epoch'] = i
best_score = metrics['score']
self.model.save(self.savpath)
print('The best metrics took place in the epoch: ',
self.metrics['epoch'])
print(
'Sensitivity: {}; Specificity: {}; Score: {}; Accuracy: {}'.format(
self.metrics['sensitivity'], self.metrics['specificity'],
self.metrics['score'], self.metrics['accuracy']))
def CreateInceptionModel(self, input_shape, classes):
# Define the input layer
X_input = Input(name='the_input', shape=input_shape)
# Stage 1 - layers before inception modules
conv1_7x7_s2 = Conv2D(filters=64,
kernel_size=(7, 7),
strides=(2, 2),
padding='same',
activation='relu',
kernel_regularizer=l2(0.01))(X_input)
maxpool1_3x3_s2 = MaxPooling2D(pool_size=(3, 3),
strides=(2, 2),
padding='same')(conv1_7x7_s2)
conv2_3x3_reduce = Conv2D(filters=64,
kernel_size=(1, 1),
padding='same',
activation='relu',
kernel_regularizer=l2(0.01))(maxpool1_3x3_s2)
conv2_3x3 = Conv2D(filters=192,
kernel_size=(3, 3),
padding='same',
activation='relu',
kernel_regularizer=l2(0.01))(conv2_3x3_reduce)
maxpool2_3x3_s2 = MaxPooling2D(pool_size=(3, 3),
strides=(2, 2),
padding='same')(conv2_3x3)
# Stage 2 - 2 inception modules and max pooling
inception_3a = inception_model(input=maxpool2_3x3_s2,
filters_1x1=64,
filters_3x3_reduce=96,
filters_3x3=128,
filters_5x5_reduce=16,
filters_5x5=32,
filters_pool_proj=32)
inception_3b = inception_model(input=inception_3a,
filters_1x1=128,
filters_3x3_reduce=128,
filters_3x3=192,
filters_5x5_reduce=32,
filters_5x5=96,
filters_pool_proj=64) #224*25*480
# Stage 3 - another type of ending layers.
# drop1 = Dropout(rate=0.4)(inception_3b)
transmute = core.Flatten()(inception_3b)
linear = Dense(units=classes,
activation='softmax',
kernel_regularizer=l2(0.01))(transmute)
last = linear
# Stage 3 - ending layers
# conv_1x1 = Conv2D(filters=classes, kernel_size=(1, 1), padding='same', activation='relu',kernel_regularizer=l2(0.01))(inception_3b)
# last = GlobalAveragePooling2D()(conv_1x1)
# last = Activation(softmax)(last)
# Create model
model = Model(inputs=X_input, outputs=last)
optimizer = optimizers.Adadelta()
model.compile(optimizer=optimizer,
loss=[focal_loss(alpha=0.25, gamma=2)])
# model.compile(optimizer=optimizer,loss='binary_crossentropy')
print('Inception neural network established. ')
return model
def CreateClassicModel(self, input_shape, classes):
X_input = Input(name='the_input', shape=input_shape)
layer_h1 = Conv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal')(X_input) # 卷积层
# layer_h1 = Dropout(0.1)(layer_h1)
layer_h2 = Conv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal')(layer_h1) # 卷积层
layer_h3 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h2) # 池化层
# layer_h3 = Dropout(0.1)(layer_h3)
layer_h4 = Conv2D(64, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal')(layer_h3) # 卷积层
# layer_h4 = Dropout(0.2)(layer_h4)
layer_h5 = Conv2D(64, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal')(layer_h4) # 卷积层
layer_h6 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h5) # 池化层
# layer_h6 = Dropout(0.2)(layer_h6)
layer_h7 = Conv2D(128, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal')(layer_h6) # 卷积层
# layer_h7 = Dropout(0.2)(layer_h7)
## layer_h8 = Conv2D(128, (3, 3), use_bias=True, activation='relu', padding='same',kernel_initializer='he_normal')(layer_h7) # 卷积层
layer_h9 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h7) # 池化层
# layer_h9 = Dropout(0.3)(layer_h9)
## layer_h10 = Conv2D(128, (3, 3), use_bias=True, activation='relu', padding='same',kernel_initializer='he_normal')(layer_h9) # 卷积层
# layer_h10 = Dropout(0.4)(layer_h10)
## layer_h11 = Conv2D(128, (3, 3), use_bias=True, activation='relu', padding='same',kernel_initializer='he_normal')(layer_h10) # 卷积层
## flayer = Flatten()(layer_h11)
flayer = Flatten()(layer_h9)
# flayer = Dropout(0.4)(flayer)
fc1 = Dense(units=32,
activation="relu",
use_bias=True,
kernel_initializer='he_normal')(flayer)
# fc1 = Dropout(0.5)(fc1)
fc2 = Dense(classes, use_bias=True,
kernel_initializer='he_normal')(fc1) # 全连接层
y_pred = Activation('softmax', name='Activation0')(fc2)
model = Model(inputs=X_input, outputs=y_pred)
optimizer = optimizers.Adadelta()
# model.compile(optimizer=optimizer, loss='binary_crossentropy')# [focal_loss])
model.compile(optimizer=optimizer,
loss=[focal_loss(alpha=0.25, gamma=2)])
return model
def CreateSimplifiedClassicModel(self, input_shape, classes):
X_input = Input(name='the_input', shape=input_shape)
layer_h1 = Conv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.01))(X_input) # 卷积层
layer_h2 = Conv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.01))(layer_h1) # 卷积层
layer_h3 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h2) # 池化层
layer_h4 = Conv2D(64, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.01))(layer_h3) # 卷积层
layer_h5 = Conv2D(64, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.01))(layer_h4) # 卷积层
layer_h6 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h5) # 池化层
layer_h7 = Conv2D(128, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.01))(layer_h6) # 卷积层
layer_h8 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h7) # 池化层
layer_h9 = Conv2D(128, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.01))(layer_h8) # 卷积层
layer_h10 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h9) # 池化层
flayer = Flatten()(layer_h10)
fc1 = Dense(units=32,
activation="relu",
use_bias=True,
kernel_initializer='he_normal')(flayer)
fc2 = Dense(classes, use_bias=True,
kernel_initializer='he_normal')(fc1) # 全连接层
y_pred = Activation('softmax', name='Activation0')(fc2)
model = Model(inputs=X_input, outputs=y_pred)
optimizer = optimizers.Adadelta()
# model.compile(optimizer=optimizer, loss='binary_crossentropy')# [focal_loss])
model.compile(optimizer=optimizer,
loss=[focal_loss(alpha=0.25, gamma=2)])
print('Simplified cnn model estabished.')
return model
def CreateSimplifiedIntensifiedClassicModel(self, input_shape, classes):
X_input = Input(name='the_input', shape=input_shape)
# temp = BatchNormalization()(X_input)
layer_h1 = Conv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=None)(X_input) # 卷积层
layer_h2 = Conv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=None)(layer_h1) # 卷积层
layer_p2 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h2) # 池化层
layer_h3 = Conv2D(64, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=None)(layer_p2) # 卷积层
layer_h4 = Conv2D(64, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=None)(layer_h3) # 卷积层
layer_p4 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h4) # 池化层
layer_h5 = Conv2D(128, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=None)(layer_p4) # 卷积层
layer_h6 = Conv2D(128, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=None)(layer_h5) # 卷积层
layer_p6 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h6) # 池化层
layer_h7 = Conv2D(256, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=None)(layer_p6) # 卷积层
layer_h8 = Conv2D(256, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=None)(layer_h7) # 卷积层
layer_p8 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h8) # 池化层
flayer = Flatten()(layer_p8)
fc2 = Dense(classes, use_bias=True,
kernel_initializer='he_normal')(flayer) # 全连接层
y_pred = Activation('softmax', name='Activation0')(fc2)
model = Model(inputs=X_input, outputs=y_pred)
optimizer = optimizers.Adadelta()
# sgd = optimizers.SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True)
# model.compile(optimizer=optimizer, loss='categorical_crossentropy')# [focal_loss])
model.compile(optimizer=optimizer,
loss=[focal_loss(alpha=0.25, gamma=2)])
print('Simplified and intensified cnn model estabished.')
self.modelname = 'cnn+dnn'
self.intelayer = Model(inputs=X_input, outputs=fc2)
return model
def TrainModel(self,
datapath,
epoch=2,
batch_size=32,
load_weights=False,
filename='model_set/speech_model25'):
assert (batch_size % CLASS_NUM == 0)
data = DataSpeech(datapath, 'train')
num_data = sum(data.DataNum) # 获取数据的数量
os.system('pkill tensorboard')
os.system('rm -rf ./checkpoints/files_summary/* ')
train_writter = tf.summary.FileWriter(
os.path.join(os.getcwd(), 'checkpoints', 'files_summary'))
os.system(
'tensorboard --logdir=/home/zhaok14/example/PycharmProjects/setsail/individual_spp/checkpoints/files_summary/ &'
)
print('\n')
print(90 * '*')
print(90 * '*')
iterations_per_epoch = min(
data.DataNum) // (batch_size // CLASS_NUM) + 1
# iterations_per_epoch = 0
print('trainer info:')
print('training data size: %d' % num_data)
print('increased epoches: ', epoch)
print('minibatch size: %d' % batch_size)
print('iterations per epoch: %d' % iterations_per_epoch)
with k.get_session() as sess:
train_writter.add_graph(sess.graph)
if load_weights == True:
try:
# modelpath = os.path.join(os.getcwd(), 'network&&weights', 'spectrogram', 'inception','spec_inception.h5')
self.model = load_model(modelpath,
custom_objects={
'focal_loss': focal_loss,
'focal_loss_fixed':
focal_loss()
})
print('Successfully loading the model.')
except:
print('Loading weights failed. Train from scratch.')
sess.run(tf.global_variables_initializer())
best_score = 0
# epoch = 0
duration = 0
for i in range(0, epoch):
iteration = 0
yielddatas = data.data_genetator(batch_size, epoch)
pbar = tqdm(yielddatas)
for input, labels in pbar:
stime = time.time()
loss = self.model.train_on_batch(input[0], labels)
temp = self.model.predict_on_batch(input[0])
dtime = time.time() - stime
duration = duration + dtime
# okay = self.model.predict_on_batch(input[0])
# compare = self.interlayer.predict_on_batch(input[0])
train_summary = tf.Summary()
train_summary.value.add(tag='loss', simple_value=loss)
train_writter.add_summary(
train_summary, iteration + i * iterations_per_epoch)
pr = 'epoch:%d/%d,iteration: %d/%d ,loss: %s' % (
epoch, i, iterations_per_epoch, iteration, loss)
pbar.set_description(pr)
if iteration == iterations_per_epoch:
break
else:
iteration += 1
pbar.close()
if i % 1 == 0:
tmetrics = self.TestModel(sess=sess,
datapath=datapath,
str_dataset='train',
data_count=1000,
out_report=False,
writer=train_writter,
step=i)
metrics = self.TestModel(sess=sess,
datapath=datapath,
str_dataset='eval',
data_count=-1,
out_report=False,
writer=train_writter,
step=i)
if PRIOR_ART == False:
condition = tmetrics['score'] > metrics[
'score'] and i > 0 and tmetrics[
'sensitivity'] >= 91 and tmetrics[
'specificity'] >= 91
else:
condition = i > 0
if condition:
if metrics['score'] >= best_score:
self.metrics = metrics
self.metrics['epoch'] = i
best_score = metrics['score']
self.savpath = []
self.savpath.append((self.baseSavPath[0] +
'_epoch' + str(i) + '.h5'))
self.savpath.append((self.baseSavPath[1] +
'_epoch' + str(i) + '.h5'))
self.model.save(self.savpath[0])
self.model.save_weights(self.savpath[1])
if 'epoch' in self.metrics.keys():
print(
'The best metric after restriction took place in the epoch: ',
self.metrics['epoch'])
print(
'Sensitivity: {}; Specificity: {}; Score: {}; Accuracy: {}'
.format(self.metrics['sensitivity'],
self.metrics['specificity'], self.metrics['score'],
self.metrics['accuracy']))
self.TestGenerability(weightspath=self.savpath[1])
else:
print('The restricted best metric is not found. Done!')
# path_test = '/home/zhaok14/example/PycharmProjects/setsail/individual_spp/network&&weights/spectrogram/mlp/spec_mlp_weights_epoch12.h5'
# self.TestGenerability(weightspath=path_test)
print('Training duration: {}s'.format(round(duration, 2)))
def CreatePureXceptionModel(self, input_shape, classes):
X_input = Input(name='the_input', shape=input_shape)
# temp = BatchNormalization()(X_input)
layer_h1 = SeparableConv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.0005))(
X_input) # 卷积层
layer_h2 = SeparableConv2D(32, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.0005))(
layer_h1) # 卷积层
layer_p2 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h2) # 池化层
layer_h3 = SeparableConv2D(64, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.0005))(
layer_p2) # 卷积层
layer_h4 = SeparableConv2D(64, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.0005))(
layer_h3) # 卷积层
layer_p4 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h4) # 池化层
layer_h5 = SeparableConv2D(128, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.0005))(
layer_p4) # 卷积层
layer_h6 = SeparableConv2D(128, (3, 3),
use_bias=True,
activation='relu',
padding='same',
kernel_initializer='he_normal',
kernel_regularizer=l2(0.0005))(
layer_h5) # 卷积层
layer_p6 = MaxPooling2D(pool_size=2, strides=None,
padding="valid")(layer_h6) # 池化层
# layer_h7 = Conv2D(256, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=None)(layer_p6) # 卷积层
# layer_h8 = Conv2D(256, (3, 3), use_bias=True, activation='relu', padding='same', kernel_initializer='he_normal',kernel_regularizer=None)(layer_h7) # 卷积层
# layer_p8 = MaxPooling2D(pool_size=2, strides=None, padding="valid")(layer_h8) # 池化层
flayer = Flatten()(layer_p6)
fc2 = Dense(classes, use_bias=True,
kernel_initializer='he_normal')(flayer) # 全连接层
y_pred = Activation('softmax', name='Activation0')(fc2)
model = Model(inputs=X_input, outputs=y_pred)
optimizer = optimizers.Adadelta()
model.compile(optimizer=optimizer,
loss=[focal_loss(alpha=0.25, gamma=2)])
print('Pure Xception cnn model estabished.')
self.modelname = 'inception'
self.intelayer = Model(inputs=X_input, outputs=fc2)
return model
