def plain_cnn():
from src.model.plain_cnn import ConvNet
convnet = ConvNet(n_channel=3, n_classes=10, image_size=24)
# convnet.debug()
convnet.train(dataloader=cifar10,
backup_path='backup/cifar10-v16/',
batch_size=128,
n_epoch=500)
def residual_net():
from src.model.residual_net import ConvNet
convnet = ConvNet(n_channel=3, n_classes=10, image_size=24, n_layers=20)
# convnet.debug()
convnet.train(dataloader=cifar10,
backup_path='backup/cifar10-v20/',
batch_size=128,
n_epoch=500)
def resnet():
from src.model.resnet import ConvNet
convnet = ConvNet(n_channel=3,
n_classes=10,
image_size=24,
network_path='src/config/networks/resnet.yaml')
convnet.train(dataloader=cifar10,
backup_path='backups/cifar10-v5/',
batch_size=128,
n_epoch=500)
def vgg_cnn():
from src.model.basic_cnn import ConvNet
convnet = ConvNet(n_channel=3,
n_classes=10,
image_size=24,
network_path='src/config/networks/vgg.yaml')
# convnet.debug()
convnet.train(dataloader=cifar10,
backup_path='backups/cifar10-v2/',
batch_size=128,
n_epoch=200)
def basic_cnn(n_channel, n_classes, batch_size, image_height, image_width,
sentence_length, vocab_size, embedding_dim, LSTM_hidden_size,
model_path, n_load_epoch):
from src.model.basic_cnn import ConvNet
convnet = ConvNet(n_channel=n_channel,
n_classes=n_classes,
image_height=image_height,
image_width=image_width)
# convnet.debug()
convnet.train(dataloader=dataloader,
backup_path=os.path.join(dir, start_time),
batch_size=batch_size,
n_epoch=n_epoch,
model_path=model_path,
n_load_epoch=n_load_epoch)
def my_v6_plain_TW():
from src.model.my_v6_plain_double import ConvNet
from src.model.my_v6_plain_double import train_obeject
# 构建会话
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
train_setting = namedtuple('train_setting', [
'only_test_small_part_dataset',
'test_proprotion',
'start_n_epoch',
'batch_size',
'n_epoch',
'backup_path',
])
train_setting = train_setting(only_test_small_part_dataset=True,
start_n_epoch=0,
test_proprotion=0.94,
batch_size=64,
n_epoch=2,
backup_path='backups/my_v6_plain_TW/')
net_dict_first = ConvNet(network_path='src/config/networks/basic.yaml',
n_channel=3,
n_classes=10,
image_size=24,
name_scope='first')
train_obeject(dataloader=cifar10,
setting=train_setting,
net_dict=net_dict_first,
sess=sess)
net_dict_second = ConvNet(network_path='src/config/networks/basic.yaml',
n_channel=3,
n_classes=10,
image_size=24,
name_scope='second')
train_obeject(dataloader=cifar10,
setting=train_setting,
net_dict=net_dict_second,
sess=sess)
tf.summary.FileWriter(train_setting.backup_path, sess.graph)
sess.close()
def my_v2_plain_TW():
from src.model.my_v2_plain_TW import ConvNet
setting = namedtuple('setting', [
'debug_mode', 'only_test_small_part_dataset', 'test_proprotion',
'start_n_epoch', 'batch_size'
])
setting = setting(debug_mode=False,
only_test_small_part_dataset=False,
start_n_epoch=0,
test_proprotion=0.94,
batch_size=64)
convnet = ConvNet(n_channel=3,
n_classes=10,
image_size=24,
network_path='src/config/networks/basic.yaml',
setting=setting)
convnet.train(dataloader=cifar10,
backup_path='backups/my_v2_plain_TW/',
n_epoch=500)
def my_v4_plain_TW_gen_one_layer():
from src.model.my_v4_plain_TW_gen_one_layer import ConvNet
setting = namedtuple('setting', [
'debug_mode', 'only_test_small_part_dataset', 'test_proprotion',
'start_n_epoch', 'batch_size', 'output_graph'
])
setting = setting(debug_mode=True,
only_test_small_part_dataset=True,
start_n_epoch=0,
test_proprotion=0.94,
batch_size=64,
output_graph=True)
convnet = ConvNet(dataloader=cifar10,
n_channel=3,
n_classes=10,
image_size=24,
network_path='src/config/networks/basic.yaml',
setting=setting)
convnet.train(backup_path='backups/my_v4_plain_TW_gen_one_layer/',
n_epoch=500)
def my_v7_plain_double_concact_to_one():
from src.model.my_v7_plain_double_concact_to_one import ConvNet
from src.model.my_v7_plain_double_concact_to_one import Conv_new_Net
from src.model.my_v7_plain_double_concact_to_one import train_obeject
from src.model.my_v7_plain_double_concact_to_one import merge_weight
# 构建会话
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
train_setting = namedtuple('train_setting', [
'only_test_small_part_dataset',
'test_proprotion',
'start_n_epoch',
'batch_size',
'n_epoch',
'backup_path',
])
train_setting = train_setting(
only_test_small_part_dataset=True,
start_n_epoch=0,
test_proprotion=0.97,
batch_size=64,
n_epoch=1,
backup_path='backups/my_v7_plain_double_concact_to_one/')
net_dict_first = ConvNet(network_path='src/config/networks/basic.yaml',
n_channel=3,
n_classes=10,
image_size=24,
name_scope='first')
train_obeject(dataloader=cifar10,
setting=train_setting,
net_dict=net_dict_first,
sess=sess)
net_dict_second = ConvNet(network_path='src/config/networks/basic.yaml',
n_channel=3,
n_classes=10,
image_size=24,
name_scope='second')
train_obeject(dataloader=cifar10,
setting=train_setting,
net_dict=net_dict_second,
sess=sess)
# 取出权重合并
var = tf.global_variables()
first_weight = [
val for val in var if 'kernel:0' in val.name and 'first' in val.name
and 'conv' in val.name
]
second_weight = [
val for val in var if 'kernel:0' in val.name and 'second' in val.name
and 'conv' in val.name
]
weight_dict = merge_weight(first_weight, second_weight)
# for name,value in weight_dict.items():
# print(name)
# print(sess.run(value))
net_dict_first = Conv_new_Net(
sess=sess,
network_path='src/config/networks/basic.yaml',
n_channel=3,
n_classes=10,
image_size=24,
name_scope='merge_net',
weight_dict=weight_dict)
train_obeject(dataloader=cifar10,
setting=train_setting,
net_dict=net_dict_first,
sess=sess)
tf.summary.FileWriter(train_setting.backup_path, sess.graph)
sess.close()
