本文整理汇总了Python中keras.callbacks.CSVLogger方法的典型用法代码示例。如果您正苦于以下问题&＃xff1a;Python callbacks.CSVLogger方法的具体用法&＃xff1f;Python callbacks.CSVLogger怎么用&＃xff1f;Python callbacks.CSVLogger使用的例子&＃xff1f;那么恭喜您, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在模块keras.callbacks的用法示例。
在下文中一共展示了callbacks.CSVLogger方法的22个代码示例&＃xff0c;这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞&＃xff0c;您的评价将有助于我们的系统推荐出更棒的Python代码示例。
示例1: resume_train
点赞 6
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def resume_train(self, category, pretrainModel, modelName, initEpoch, batchSize&＃61;8, epochs&＃61;20):
self.modelName &＃61; modelName
self.load_model(pretrainModel)
refineNetflag &＃61; True
self.nStackNum &＃61; 2
modelPath &＃61; os.path.dirname(pretrainModel)
trainDt &＃61; DataGenerator(category, os.path.join("../../data/train/Annotations", "train_split.csv"))
trainGen &＃61; trainDt.generator_with_mask_ohem(graph&＃61;tf.get_default_graph(), kerasModel&＃61;self.model,
batchSize&＃61;batchSize, inputSize&＃61;(self.inputHeight, self.inputWidth),
nStackNum&＃61;self.nStackNum, flipFlag&＃61;False, cropFlag&＃61;False)
normalizedErrorCallBack &＃61; NormalizedErrorCallBack("../../trained_models/", category, refineNetflag, resumeFolder&＃61;modelPath)
csvlogger &＃61; CSVLogger(os.path.join(normalizedErrorCallBack.get_folder_path(),
"csv_train_" &＃43; self.modelName &＃43; "_" &＃43; str(
datetime.datetime.now().strftime(&＃39;%H:%M&＃39;)) &＃43; ".csv"))
self.model.fit_generator(initial_epoch&＃61;initEpoch, generator&＃61;trainGen, steps_per_epoch&＃61;trainDt.get_dataset_size() // batchSize,
epochs&＃61;epochs, callbacks&＃61;[normalizedErrorCallBack, csvlogger])
开发者ID:yuanyuanli85&＃xff0c;项目名称:FashionAI_KeyPoint_Detection_Challenge_Keras&＃xff0c;代码行数:24&＃xff0c;
示例2: train
点赞 6
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train():
data &＃61; load_train_data()
data &＃61; data.reshape((data.shape[0],data.shape[1],data.shape[2],1))
data &＃61; data.astype(&＃39;float32&＃39;)/255.0
# model selection
if args.pretrain: model &＃61; load_model(args.pretrain, compile&＃61;False)
else:
if args.model &＃61;&＃61; &＃39;DnCNN&＃39;: model &＃61; models.DnCNN()
# compile the model
model.compile(optimizer&＃61;Adam(), loss&＃61;[&＃39;mse&＃39;])
# use call back functions
ckpt &＃61; ModelCheckpoint(save_dir&＃43;&＃39;/model_{epoch:02d}.h5&＃39;, monitor&＃61;&＃39;val_loss&＃39;,
verbose&＃61;0, period&＃61;args.save_every)
csv_logger &＃61; CSVLogger(save_dir&＃43;&＃39;/log.csv&＃39;, append&＃61;True, separator&＃61;&＃39;,&＃39;)
lr &＃61; LearningRateScheduler(step_decay)
# train
history &＃61; model.fit_generator(train_datagen(data, batch_size&＃61;args.batch_size),
steps_per_epoch&＃61;len(data)//args.batch_size, epochs&＃61;args.epoch, verbose&＃61;1,
callbacks&＃61;[ckpt, csv_logger, lr])
return model
开发者ID:husqin&＃xff0c;项目名称:DnCNN-keras&＃xff0c;代码行数:25&＃xff0c;
示例3: get_callbacks
点赞 6
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def get_callbacks(model_file, initial_learning_rate&＃61;0.0001, learning_rate_drop&＃61;0.5, learning_rate_epochs&＃61;None,
learning_rate_patience&＃61;50, logging_file&＃61;"training.log", verbosity&＃61;1,
early_stopping_patience&＃61;None):
callbacks &＃61; list()
callbacks.append(ModelCheckpoint(model_file,monitor&＃61;&＃39;val_acc&＃39;, save_best_only&＃61;True,verbose&＃61;verbosity, save_weights_only&＃61;True))
# callbacks.append(ModelCheckpoint(model_file, save_best_only&＃61;True, save_weights_only&＃61;True))
callbacks.append(CSVLogger(logging_file, append&＃61;True))
if learning_rate_epochs:
callbacks.append(LearningRateScheduler(partial(step_decay, initial_lrate&＃61;initial_learning_rate,
drop&＃61;learning_rate_drop, epochs_drop&＃61;learning_rate_epochs)))
else:
callbacks.append(ReduceLROnPlateau(factor&＃61;learning_rate_drop, patience&＃61;learning_rate_patience,
verbose&＃61;verbosity))
if early_stopping_patience:
callbacks.append(EarlyStopping(verbose&＃61;verbosity, patience&＃61;early_stopping_patience))
return callbacks
开发者ID:wcfzl&＃xff0c;项目名称:3D-CNNs-for-Liver-Classification&＃xff0c;代码行数:18&＃xff0c;
示例4: main
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def main(rootdir, case, results):
train_x, train_y, valid_x, valid_y, test_x, test_y &＃61; get_data(args.dataset, case)
input_shape &＃61; (train_x.shape[1], train_x.shape[2])
num_class &＃61; train_y.shape[1]
if not os.path.exists(rootdir):
os.makedirs(rootdir)
filepath &＃61; os.path.join(rootdir, str(case) &＃43; &＃39;.hdf5&＃39;)
saveto &＃61; os.path.join(rootdir, str(case) &＃43; &＃39;.csv&＃39;)
optimizer &＃61; Adam(lr&＃61;args.lr, clipnorm&＃61;args.clip)
pred_dir &＃61; os.path.join(rootdir, str(case) &＃43; &＃39;_pred.txt&＃39;)
if args.train:
model &＃61; creat_model(input_shape, num_class)
early_stop &＃61; EarlyStopping(monitor&＃61;&＃39;val_acc&＃39;, patience&＃61;15, mode&＃61;&＃39;auto&＃39;)
reduce_lr &＃61; ReduceLROnPlateau(monitor&＃61;&＃39;val_acc&＃39;, factor&＃61;0.1, patience&＃61;5, mode&＃61;&＃39;auto&＃39;, cooldown&＃61;3., verbose&＃61;1)
checkpoint &＃61; ModelCheckpoint(filepath, monitor&＃61;&＃39;val_acc&＃39;, verbose&＃61;1, save_best_only&＃61;True, mode&＃61;&＃39;auto&＃39;)
csv_logger &＃61; CSVLogger(saveto)
if args.dataset&＃61;&＃61;&＃39;NTU&＃39; or args.dataset &＃61;&＃61; &＃39;PKU&＃39;:
callbacks_list &＃61; [csv_logger, checkpoint, early_stop, reduce_lr]
else:
callbacks_list &＃61; [csv_logger, checkpoint]
model.compile(loss&＃61;&＃39;categorical_crossentropy&＃39;, optimizer&＃61;optimizer, metrics&＃61;[&＃39;accuracy&＃39;])
model.fit(train_x, train_y, validation_data&＃61;[valid_x, valid_y], epochs&＃61;args.epochs,
batch_size&＃61;args.batch_size, callbacks&＃61;callbacks_list, verbose&＃61;2)
# test
model &＃61; creat_model(input_shape, num_class)
model.load_weights(filepath)
model.compile(loss&＃61;&＃39;categorical_crossentropy&＃39;, optimizer&＃61;optimizer, metrics&＃61;[&＃39;accuracy&＃39;])
scores &＃61; get_activation(model, test_x, test_y, pred_dir, VA&＃61;10, par&＃61;9)
results.append(round(scores, 2))
开发者ID:microsoft&＃xff0c;项目名称:View-Adaptive-Neural-Networks-for-Skeleton-based-Human-Action-Recognition&＃xff0c;代码行数:37&＃xff0c;
示例5: train
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train(model, data, args):
"""
Training a CapsuleNet
:param model: the CapsuleNet model
:param data: a tuple containing training and testing data, like &＃96;((x_train, y_train), (x_test, y_test))&＃96;
:param args: arguments
:return: The trained model
"""
(x_train, y_train), (x_test, y_test) &＃61; data
log &＃61; callbacks.CSVLogger(args.save_dir &＃43; &＃39;/log.csv&＃39;)
checkpoint &＃61; callbacks.ModelCheckpoint(args.save_dir &＃43; &＃39;/weights-{epoch:02d}.h5&＃39;, monitor&＃61;&＃39;val_capsnet_acc&＃39;,
save_best_only&＃61;False, save_weights_only&＃61;True, verbose&＃61;1)
lr_decay &＃61; callbacks.LearningRateScheduler(schedule&＃61;lambda epoch: args.lr * (args.lr_decay ** epoch))
model.compile(optimizer&＃61;optimizers.Adam(lr&＃61;args.lr),
loss&＃61;[margin_loss, &＃39;mse&＃39;],
loss_weights&＃61;[1., args.lam_recon],
metrics&＃61;{&＃39;capsnet&＃39;: &＃39;accuracy&＃39;})
def train_generator(x, y, batch_size, shift_fraction&＃61;0.):
train_datagen &＃61; ImageDataGenerator(width_shift_range&＃61;shift_fraction,
height_shift_range&＃61;shift_fraction)
generator &＃61; train_datagen.flow(x, y, batch_size&＃61;batch_size)
while 1:
x_batch, y_batch &＃61; generator.next()
yield ([x_batch, y_batch], [y_batch, x_batch])
model.fit_generator(generator&＃61;train_generator(x_train, y_train, args.batch_size, args.shift_fraction),
steps_per_epoch&＃61;int(y_train.shape[0] / args.batch_size),
epochs&＃61;args.epochs,
shuffle &＃61; True,
validation_data&＃61;[[x_test, y_test], [y_test, x_test]],
callbacks&＃61;snapshot.get_callbacks(log,model_prefix&＃61;model_prefix))
model.save_weights(args.save_dir &＃43; &＃39;/trained_model.h5&＃39;)
print(&＃39;Trained model saved to \&＃39;%s/trained_model.h5\&＃39;&＃39; % args.save_dir)
return model
开发者ID:vinojjayasundara&＃xff0c;项目名称:textcaps&＃xff0c;代码行数:41&＃xff0c;
示例6: train
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train(train_set, val_set, cfg, config_name, resume, model_path):
if not(model_path is None):
if resume:
print("Loading compiled model: " &＃43; model_path)
model &＃61; keras.models.load_model(model_path, compile&＃61;True)
else:
print("Loading uncompiled model: " &＃43; model_path)
model &＃61; keras.models.load_model(model_path, compile&＃61;False)
model &＃61; compile_model(model, cfg["model"])
else:
print("Loading the network..")
model &＃61; load_model(cfg["model"])
csv_logger &＃61; CSVLogger(&＃39;checkpoint/&＃39; &＃43; config_name &＃43;
&＃39;-training.log&＃39;, append&＃61;resume)
save_ckpt &＃61; ModelCheckpoint("checkpoint/weights.{epoch:02d}-{val_loss:.2f}" &＃43; config_name &＃43; ".hdf5", monitor&＃61;&＃39;val_loss&＃39;,
verbose&＃61;1,
save_best_only&＃61;True,
period&＃61;1)
early_stopping &＃61; EarlyStopping(monitor&＃61;&＃39;val_loss&＃39;,
min_delta&＃61;0,
patience&＃61;5,
verbose&＃61;0, mode&＃61;&＃39;auto&＃39;)
lr_schedule &＃61; ReduceLROnPlateau(
monitor&＃61;&＃39;val_loss&＃39;, factor&＃61;0.1, patience&＃61;3, verbose&＃61;1, mode&＃61;&＃39;auto&＃39;, min_lr&＃61;10e-7)
callback_list &＃61; [save_ckpt, early_stopping, lr_schedule, csv_logger]
print("Start the training..")
model.fit_generator(train_set,
epochs&＃61;cfg["nb_epoch"],
callbacks&＃61;callback_list,
validation_data&＃61;val_set,
workers&＃61;cfg["workers"],
use_multiprocessing&＃61;cfg["use_multiprocessing"],
shuffle&＃61;True
)
开发者ID:qlemaire22&＃xff0c;项目名称:speech-music-detection&＃xff0c;代码行数:43&＃xff0c;
示例7: train
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train(self, category, batchSize&＃61;8, epochs&＃61;20, lrschedule&＃61;False):
trainDt &＃61; DataGenerator(category, os.path.join("../../data/train/Annotations", "train_split.csv"))
trainGen &＃61; trainDt.generator_with_mask_ohem( graph&＃61;tf.get_default_graph(), kerasModel&＃61;self.model,
batchSize&＃61; batchSize, inputSize&＃61;(self.inputHeight, self.inputWidth),
nStackNum&＃61;self.nStackNum, flipFlag&＃61;False, cropFlag&＃61;False)
normalizedErrorCallBack &＃61; NormalizedErrorCallBack("../../trained_models/", category, True)
csvlogger &＃61; CSVLogger( os.path.join(normalizedErrorCallBack.get_folder_path(),
"csv_train_"&＃43;self.modelName&＃43;"_"&＃43;str(datetime.datetime.now().strftime(&＃39;%H:%M&＃39;))&＃43;".csv"))
xcallbacks &＃61; [normalizedErrorCallBack, csvlogger]
self.model.fit_generator(generator&＃61;trainGen, steps_per_epoch&＃61;trainDt.get_dataset_size()//batchSize,
epochs&＃61;epochs, callbacks&＃61;xcallbacks)
开发者ID:yuanyuanli85&＃xff0c;项目名称:FashionAI_KeyPoint_Detection_Challenge_Keras&＃xff0c;代码行数:17&＃xff0c;
示例8: train_gan
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train_gan( dataf ) :
gen, disc, gan &＃61; build_networks()
# Uncomment these, if you want to continue training from some snapshot.
# (or load pretrained generator weights)
#load_weights(gen, Args.genw)
#load_weights(disc, Args.discw)
logger &＃61; CSVLogger(&＃39;loss.csv&＃39;) # yeah, you can use callbacks independently
logger.on_train_begin() # initialize csv file
with h5py.File( dataf, &＃39;r&＃39; ) as f :
faces &＃61; f.get( &＃39;faces&＃39; )
run_batches(gen, disc, gan, faces, logger, range(5000))
logger.on_train_end()
开发者ID:forcecore&＃xff0c;项目名称:Keras-GAN-Animeface-Character&＃xff0c;代码行数:16&＃xff0c;
示例9: train_model
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train_model(self, train_generator, steps_per_epoch&＃61;None, epochs&＃61;1, validation_generator&＃61;None,
validation_steps&＃61;None, workers&＃61;1, use_multiprocessing&＃61;False, shuffle&＃61;True, initial_epoch&＃61;0,
save_history&＃61;False, save_model_per_epoch&＃61;False):
saved_items_dir &＃61; os.path.join(os.path.dirname(__file__), os.pardir, &＃39;saved_items&＃39;)
if not os.path.exists(saved_items_dir):
os.makedirs(saved_items_dir)
callbacks &＃61; []
if save_history:
history_file &＃61; os.path.join(saved_items_dir, &＃39;history&＃39;)
csv_logger &＃61; CSVLogger(history_file, append&＃61;True)
callbacks.append(csv_logger)
if save_model_per_epoch:
save_model_file &＃61; os.path.join(saved_items_dir, &＃39;bidaf_{epoch:02d}.h5&＃39;)
checkpointer &＃61; ModelCheckpoint(filepath&＃61;save_model_file, verbose&＃61;1)
callbacks.append(checkpointer)
history &＃61; self.model.fit_generator(train_generator, steps_per_epoch&＃61;steps_per_epoch, epochs&＃61;epochs,
callbacks&＃61;callbacks, validation_data&＃61;validation_generator,
validation_steps&＃61;validation_steps, workers&＃61;workers,
use_multiprocessing&＃61;use_multiprocessing, shuffle&＃61;shuffle,
initial_epoch&＃61;initial_epoch)
if not save_model_per_epoch:
self.model.save(os.path.join(saved_items_dir, &＃39;bidaf.h5&＃39;))
return history, self.model
开发者ID:ParikhKadam&＃xff0c;项目名称:bidaf-keras&＃xff0c;代码行数:31&＃xff0c;
示例10: pretrain
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def pretrain(self, x, y&＃61;None, optimizer&＃61;&＃39;adam&＃39;, epochs&＃61;200, batch_size&＃61;256, save_dir&＃61;&＃39;results/temp&＃39;):
print(&＃39;...Pretraining...&＃39;)
self.autoencoder.compile(optimizer&＃61;optimizer, loss&＃61;&＃39;mse&＃39;)
csv_logger &＃61; callbacks.CSVLogger(save_dir &＃43; &＃39;/pretrain_log.csv&＃39;)
cb &＃61; [csv_logger]
if y is not None:
class PrintACC(callbacks.Callback):
def __init__(self, x, y):
self.x &＃61; x
self.y &＃61; y
super(PrintACC, self).__init__()
def on_epoch_end(self, epoch, logs&＃61;None):
if int(epochs/10) !&＃61; 0 and epoch % int(epochs/10) !&＃61; 0:
return
feature_model &＃61; Model(self.model.input,
self.model.get_layer(
&＃39;encoder_%d&＃39; % (int(len(self.model.layers) / 2) - 1)).output)
features &＃61; feature_model.predict(self.x)
km &＃61; KMeans(n_clusters&＃61;len(np.unique(self.y)), n_init&＃61;20, n_jobs&＃61;4)
y_pred &＃61; km.fit_predict(features)
# print()
print(&＃39; &＃39;*8 &＃43; &＃39;|&＃61;&＃61;> acc: %.4f, nmi: %.4f <&＃61;&＃61;|&＃39;
% (metrics.acc(self.y, y_pred), metrics.nmi(self.y, y_pred)))
cb.append(PrintACC(x, y))
# begin pretraining
t0 &＃61; time()
self.autoencoder.fit(x, x, batch_size&＃61;batch_size, epochs&＃61;epochs, callbacks&＃61;cb)
print(&＃39;Pretraining time: %ds&＃39; % round(time() - t0))
self.autoencoder.save_weights(save_dir &＃43; &＃39;/ae_weights.h5&＃39;)
print(&＃39;Pretrained weights are saved to %s/ae_weights.h5&＃39; % save_dir)
self.pretrained &＃61; True
开发者ID:XifengGuo&＃xff0c;项目名称:DEC-keras&＃xff0c;代码行数:37&＃xff0c;
示例11: get_callbacks
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def get_callbacks(model_file, initial_learning_rate&＃61;0.0001, learning_rate_drop&＃61;0.5, learning_rate_epochs&＃61;None,
learning_rate_patience&＃61;50, logging_file&＃61;"training.log", verbosity&＃61;1,
early_stopping_patience&＃61;None):
callbacks &＃61; list()
callbacks.append(ModelCheckpoint(model_file, save_best_only&＃61;True))
callbacks.append(CSVLogger(logging_file, append&＃61;True))
if learning_rate_epochs:
callbacks.append(LearningRateScheduler(partial(step_decay, initial_lrate&＃61;initial_learning_rate,
drop&＃61;learning_rate_drop, epochs_drop&＃61;learning_rate_epochs)))
else:
callbacks.append(ReduceLROnPlateau(factor&＃61;learning_rate_drop, patience&＃61;learning_rate_patience,
verbose&＃61;verbosity))
if early_stopping_patience:
callbacks.append(EarlyStopping(verbose&＃61;verbosity, patience&＃61;early_stopping_patience))
return callbacks
开发者ID:ellisdg&＃xff0c;项目名称:3DUnetCNN&＃xff0c;代码行数:17&＃xff0c;
示例12: train
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train(model, epochs, patience, output_path, nproc, train_obj, val_obj):
"""
:param model: model to train (must be compiled)
:type model: Model
:param epochs: max number of epochs to train.
:type epochs: int
:param patience: Stop after these many layers if val. loss doesn&＃39;t decrease
:type patience: int
:param output_path: paths to save weights and logs
:type output_path: str
:param nproc: number of processors for training
:type nproc: int
:param train_obj: DataGenerator training object for training
:type train_obj: DataGenerator
:param val_obj: DataGenerator training object for validation
:type val_obj: DataGenerator
:return: model, history object
"""
if nproc &＃61;&＃61; 1:
use_multiprocessing &＃61; False
else:
use_multiprocessing &＃61; True
# Callbacks for training and validation
ES &＃61; EarlyStopping(monitor&＃61;&＃39;val_loss&＃39;, min_delta&＃61;1e-3, patience&＃61;patience, verbose&＃61;1, mode&＃61;&＃39;min&＃39;,
restore_best_weights&＃61;True)
CK &＃61; ModelCheckpoint(output_path &＃43; &＃39;weights.h5&＃39;, monitor&＃61;&＃39;val_loss&＃39;, verbose&＃61;1, save_best_only&＃61;True,
save_weights_only&＃61;False,
mode&＃61;&＃39;min&＃39;)
csv_name &＃61; output_path &＃43; &＃39;training_log.csv&＃39;
LO &＃61; CSVLogger(csv_name, append&＃61;False)
callbacks &＃61; [ES, CK, LO]
train_history &＃61; model.fit_generator(generator&＃61;train_obj, validation_data&＃61;val_obj, epochs&＃61;epochs,
use_multiprocessing&＃61;use_multiprocessing, max_queue_size&＃61;10, workers&＃61;nproc,
shuffle&＃61;True, callbacks&＃61;callbacks, verbose&＃61;1)
return model, train_history
开发者ID:devanshkv&＃xff0c;项目名称:fetch&＃xff0c;代码行数:41&＃xff0c;
示例13: initialize_parameters
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def initialize_parameters():
mnist_common &＃61; mnist.MNIST(mnist.file_path,
&＃39;mnist_params.txt&＃39;,
&＃39;keras&＃39;,
prog&＃61;&＃39;mnist_mlp&＃39;,
desc&＃61;&＃39;MNIST example&＃39;
)
# Initialize parameters
gParameters &＃61; candle.finalize_parameters(mnist_common)
csv_logger &＃61; CSVLogger(&＃39;{}/params.log&＃39;.format(gParameters))
return gParameters
开发者ID:ECP-CANDLE&＃xff0c;项目名称:Benchmarks&＃xff0c;代码行数:15&＃xff0c;
示例14: initialize_parameters
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def initialize_parameters():
mnist_common &＃61; mnist.MNIST(mnist.file_path,
&＃39;mnist_params.txt&＃39;,
&＃39;keras&＃39;,
prog&＃61;&＃39;mnist_cnn&＃39;,
desc&＃61;&＃39;MNIST CNN example&＃39;
)
# Initialize parameters
gParameters &＃61; candle.finalize_parameters(mnist_common)
csv_logger &＃61; CSVLogger(&＃39;{}/params.log&＃39;.format(gParameters))
return gParameters
开发者ID:ECP-CANDLE&＃xff0c;项目名称:Benchmarks&＃xff0c;代码行数:15&＃xff0c;
示例15: get_callbacks
点赞 5
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def get_callbacks(arguments):
if arguments.net.find(&＃39;caps&＃39;) !&＃61; -1:
monitor_name &＃61; &＃39;val_out_seg_dice_hard&＃39;
else:
monitor_name &＃61; &＃39;val_dice_hard&＃39;
csv_logger &＃61; CSVLogger(join(arguments.log_dir, arguments.output_name &＃43; &＃39;_log_&＃39; &＃43; arguments.time &＃43; &＃39;.csv&＃39;), separator&＃61;&＃39;,&＃39;)
tb &＃61; TensorBoard(arguments.tf_log_dir, batch_size&＃61;arguments.batch_size, histogram_freq&＃61;0)
model_checkpoint &＃61; ModelCheckpoint(join(arguments.check_dir, arguments.output_name &＃43; &＃39;_model_&＃39; &＃43; arguments.time &＃43; &＃39;.hdf5&＃39;),
monitor&＃61;monitor_name, save_best_only&＃61;True, save_weights_only&＃61;True,
verbose&＃61;1, mode&＃61;&＃39;max&＃39;)
lr_reducer &＃61; ReduceLROnPlateau(monitor&＃61;monitor_name, factor&＃61;0.05, cooldown&＃61;0, patience&＃61;5,verbose&＃61;1, mode&＃61;&＃39;max&＃39;)
early_stopper &＃61; EarlyStopping(monitor&＃61;monitor_name, min_delta&＃61;0, patience&＃61;25, verbose&＃61;0, mode&＃61;&＃39;max&＃39;)
return [model_checkpoint, csv_logger, lr_reducer, early_stopper, tb]
开发者ID:lalonderodney&＃xff0c;项目名称:SegCaps&＃xff0c;代码行数:17&＃xff0c;
示例16: train_model
点赞 4
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train_model(self,file_list,labels,n_fold&＃61;5,batch_size&＃61;16,epochs&＃61;40,dim&＃61;224,lr&＃61;1e-5,model&＃61;&＃39;ResNet50&＃39;):
model_save_dest &＃61; {}
k &＃61; 0
kf &＃61; KFold(n_splits&＃61;n_fold, random_state&＃61;0, shuffle&＃61;True)
for train_index,test_index in kf.split(file_list):
k &＃43;&＃61; 1
file_list &＃61; np.array(file_list)
labels &＃61; np.array(labels)
train_files,train_labels &＃61; file_list[train_index],labels[train_index]
val_files,val_labels &＃61; file_list[test_index],labels[test_index]
if model &＃61;&＃61; &＃39;Resnet50&＃39;:
model_final &＃61; self.resnet_pseudo(dim&＃61;224,freeze_layers&＃61;10,full_freeze&＃61;&＃39;N&＃39;)
if model &＃61;&＃61; &＃39;VGG16&＃39;:
model_final &＃61; self.VGG16_pseudo(dim&＃61;224,freeze_layers&＃61;10,full_freeze&＃61;&＃39;N&＃39;)
if model &＃61;&＃61; &＃39;InceptionV3&＃39;:
model_final &＃61; self.inception_pseudo(dim&＃61;224,freeze_layers&＃61;10,full_freeze&＃61;&＃39;N&＃39;)
adam &＃61; optimizers.Adam(lr&＃61;lr, beta_1&＃61;0.9, beta_2&＃61;0.999, epsilon&＃61;1e-08, decay&＃61;0.0)
model_final.compile(optimizer&＃61;adam, loss&＃61;["mse"],metrics&＃61;[&＃39;mse&＃39;])
reduce_lr &＃61; keras.callbacks.ReduceLROnPlateau(monitor&＃61;&＃39;val_loss&＃39;, factor&＃61;0.50,patience&＃61;3, min_lr&＃61;0.000001)
early &＃61; EarlyStopping(monitor&＃61;&＃39;val_loss&＃39;, patience&＃61;10, mode&＃61;&＃39;min&＃39;, verbose&＃61;1)
logger &＃61; CSVLogger(&＃39;keras-5fold-run-01-v1-epochs_ib.log&＃39;, separator&＃61;&＃39;,&＃39;, append&＃61;False)
checkpoint &＃61; ModelCheckpoint(
&＃39;kera1-5fold-run-01-v1-fold-&＃39; &＃43; str(&＃39;%02d&＃39; % (k &＃43; 1)) &＃43; &＃39;-run-&＃39; &＃43; str(&＃39;%02d&＃39; % (1 &＃43; 1)) &＃43; &＃39;.check&＃39;,
monitor&＃61;&＃39;val_loss&＃39;, mode&＃61;&＃39;min&＃39;,
save_best_only&＃61;True,
verbose&＃61;1)
callbacks &＃61; [reduce_lr,early,checkpoint,logger]
train_gen &＃61; DataGenerator(train_files,train_labels,batch_size&＃61;32,n_classes&＃61;len(self.class_folders),dim&＃61;(self.dim,self.dim,3),shuffle&＃61;True)
val_gen &＃61; DataGenerator(val_files,val_labels,batch_size&＃61;32,n_classes&＃61;len(self.class_folders),dim&＃61;(self.dim,self.dim,3),shuffle&＃61;True)
model_final.fit_generator(train_gen,epochs&＃61;epochs,verbose&＃61;1,validation_data&＃61;(val_gen),callbacks&＃61;callbacks)
model_name &＃61; &＃39;kera1-5fold-run-01-v1-fold-&＃39; &＃43; str(&＃39;%02d&＃39; % (k &＃43; 1)) &＃43; &＃39;-run-&＃39; &＃43; str(&＃39;%02d&＃39; % (1 &＃43; 1)) &＃43; &＃39;.check&＃39;
del model_final
f &＃61; h5py.File(model_name, &＃39;r&＃43;&＃39;)
del f[&＃39;optimizer_weights&＃39;]
f.close()
model_final &＃61; keras.models.load_model(model_name)
model_name1 &＃61; self.outdir &＃43; str(model) &＃43; &＃39;___&＃39; &＃43; str(k)
model_final.save(model_name1)
model_save_dest[k] &＃61; model_name1
return model_save_dest
# Hold out dataset validation function
开发者ID:PacktPublishing&＃xff0c;项目名称:Intelligent-Projects-Using-Python&＃xff0c;代码行数:52&＃xff0c;
示例17: train_model
点赞 4
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train_model(self,train_dir,val_dir,n_fold&＃61;5,batch_size&＃61;16,epochs&＃61;40,dim&＃61;224,lr&＃61;1e-5,model&＃61;&＃39;ResNet50&＃39;):
if model &＃61;&＃61; &＃39;Resnet50&＃39;:
model_final &＃61; self.resnet_pseudo(dim&＃61;224,freeze_layers&＃61;10,full_freeze&＃61;&＃39;N&＃39;)
if model &＃61;&＃61; &＃39;VGG16&＃39;:
model_final &＃61; self.VGG16_pseudo(dim&＃61;224,freeze_layers&＃61;10,full_freeze&＃61;&＃39;N&＃39;)
if model &＃61;&＃61; &＃39;InceptionV3&＃39;:
model_final &＃61; self.inception_pseudo(dim&＃61;224,freeze_layers&＃61;10,full_freeze&＃61;&＃39;N&＃39;)
train_file_names &＃61; glob.glob(f&＃39;{train_dir}/*/*&＃39;)
val_file_names &＃61; glob.glob(f&＃39;{val_dir}/*/*&＃39;)
train_steps_per_epoch &＃61; len(train_file_names)/float(batch_size)
val_steps_per_epoch &＃61; len(val_file_names)/float(batch_size)
train_datagen &＃61; ImageDataGenerator(horizontal_flip &＃61; True,vertical_flip &＃61; True,width_shift_range &＃61; 0.1,height_shift_range &＃61; 0.1,
channel_shift_range&＃61;0,zoom_range &＃61; 0.2,rotation_range &＃61; 20,preprocessing_function&＃61;pre_process)
val_datagen &＃61; ImageDataGenerator(preprocessing_function&＃61;pre_process)
train_generator &＃61; train_datagen.flow_from_directory(train_dir,
target_size&＃61;(dim,dim),
batch_size&＃61;batch_size,
class_mode&＃61;&＃39;categorical&＃39;)
val_generator &＃61; val_datagen.flow_from_directory(val_dir,
target_size&＃61;(dim,dim),
batch_size&＃61;batch_size,
class_mode&＃61;&＃39;categorical&＃39;)
print(train_generator.class_indices)
joblib.dump(train_generator.class_indices,f&＃39;{self.outdir}/class_indices.pkl&＃39;)
adam &＃61; optimizers.Adam(lr&＃61;lr, beta_1&＃61;0.9, beta_2&＃61;0.999, epsilon&＃61;1e-08, decay&＃61;0.0)
model_final.compile(optimizer&＃61;adam, loss&＃61;["categorical_crossentropy"],metrics&＃61;[&＃39;accuracy&＃39;])
reduce_lr &＃61; keras.callbacks.ReduceLROnPlateau(monitor&＃61;&＃39;val_loss&＃39;, factor&＃61;0.50,patience&＃61;3, min_lr&＃61;0.000001)
early &＃61; EarlyStopping(monitor&＃61;&＃39;val_loss&＃39;, patience&＃61;10, mode&＃61;&＃39;min&＃39;, verbose&＃61;1)
logger &＃61; CSVLogger(f&＃39;{self.outdir}/keras-epochs_ib.log&＃39;, separator&＃61;&＃39;,&＃39;, append&＃61;False)
model_name &＃61; f&＃39;{self.outdir}/keras_transfer_learning-run.check&＃39;
checkpoint &＃61; ModelCheckpoint(
model_name,
monitor&＃61;&＃39;val_loss&＃39;, mode&＃61;&＃39;min&＃39;,
save_best_only&＃61;True,
verbose&＃61;1)
callbacks &＃61; [reduce_lr,early,checkpoint,logger]
model_final.fit_generator(train_generator,steps_per_epoch&＃61;train_steps_per_epoch,epochs&＃61;epochs,verbose&＃61;1,validation_data&＃61;(val_generator),validation_steps&＃61;val_steps_per_epoch,callbacks&＃61;callbacks,
class_weight&＃61;{0:0.012,1:0.12,2:0.058,3:0.36,4:0.43})
#model_final.fit_generator(train_generator,steps_per_epoch&＃61;1,epochs&＃61;epochs,verbose&＃61;1,validation_data&＃61;(val_generator),validation_steps&＃61;1,callbacks&＃61;callbacks)
del model_final
f &＃61; h5py.File(model_name, &＃39;r&＃43;&＃39;)
del f[&＃39;optimizer_weights&＃39;]
f.close()
model_final &＃61; keras.models.load_model(model_name)
model_to_store_path &＃61; f&＃39;{self.outdir}/{model}&＃39;
model_final.save(model_to_store_path)
return model_to_store_path,train_generator.class_indices
# Hold out dataset validation function
开发者ID:PacktPublishing&＃xff0c;项目名称:Intelligent-Projects-Using-Python&＃xff0c;代码行数:53&＃xff0c;
示例18: train
点赞 4
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train(model, data, args):
"""
Training a 3-level DCNet
:param model: the 3-level DCNet model
:param data: a tuple containing training and testing data, like &＃96;((x_train, y_train), (x_test, y_test))&＃96;
:param args: arguments
:return: The trained model
"""
# unpacking the data
(x_train, y_train), (x_test, y_test) &＃61; data
row &＃61; x_train.shape[1]
col &＃61; x_train.shape[2]
channel &＃61; x_train.shape[3]
# callbacks
log &＃61; callbacks.CSVLogger(args.save_dir &＃43; &＃39;/log.csv&＃39;)
tb &＃61; callbacks.TensorBoard(log_dir&＃61;args.save_dir &＃43; &＃39;/tensorboard-logs&＃39;, histogram_freq&＃61;int(args.debug))
checkpoint &＃61; callbacks.ModelCheckpoint(args.save_dir &＃43; &＃39;/weights-{epoch:02d}.h5&＃39;, monitor&＃61;&＃39;val_capsnet_acc&＃39;,
verbose&＃61;1)
lr_decay &＃61; callbacks.LearningRateScheduler(schedule&＃61;lambda epoch: args.lr * (args.lr_decay ** epoch))
# compile the model
# Notice the four separate losses (for separate backpropagations)
model.compile(optimizer&＃61;optimizers.Adam(lr&＃61;args.lr),
loss&＃61;[margin_loss, margin_loss, margin_loss, margin_loss, &＃39;mse&＃39;],
loss_weights&＃61;[1., 1., 1., 1., args.lam_recon],
metrics&＃61;{&＃39;capsnet&＃39;: &＃39;accuracy&＃39;})
#model.load_weights(&＃39;result/weights.h5&＃39;)
"""
# Training without data augmentation:
model.fit([x_train, y_train], [y_train, y_train, y_train, y_train, x_train], batch_size&＃61;args.batch_size, epochs&＃61;args.epochs,
validation_data&＃61;[[x_test, y_test], [y_test, y_test, y_test, y_test, x_test]], callbacks&＃61;[log, tb, checkpoint, lr_decay])
"""
# Training with data augmentation
def train_generator(x, y, batch_size, shift_fraction&＃61;0.):
train_datagen &＃61; ImageDataGenerator(width_shift_range&＃61;shift_fraction,
height_shift_range&＃61;shift_fraction) # shift up to 2 pixel for MNIST
generator &＃61; train_datagen.flow(x, y, batch_size&＃61;batch_size)
while 1:
x_batch, y_batch &＃61; generator.next()
yield ([x_batch, y_batch], [y_batch, y_batch, y_batch, y_batch, x_batch[:,:,:,0:1]])
# Training with data augmentation. If shift_fraction&＃61;0., also no augmentation.
model.fit_generator(generator&＃61;train_generator(x_train, y_train, args.batch_size, args.shift_fraction),
steps_per_epoch&＃61;int(y_train.shape[0] / args.batch_size),
epochs&＃61;args.epochs,
validation_data&＃61;[[x_test, y_test], [y_test, y_test, y_test, y_test, x_test[:,:,:,0:1]]],
callbacks&＃61;[log, tb, checkpoint, lr_decay])
# Save model weights
model.save_weights(args.save_dir &＃43; &＃39;/trained_model.h5&＃39;)
print(&＃39;Trained model saved to \&＃39;%s/trained_model.h5\&＃39;&＃39; % args.save_dir)
plot_log(args.save_dir &＃43; &＃39;/log.csv&＃39;, show&＃61;True)
return model
开发者ID:ssrp&＃xff0c;项目名称:Multi-level-DCNet&＃xff0c;代码行数:62&＃xff0c;
示例19: train
点赞 4
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train(model, data, args):
"""
Training a 3-level DCNet
:param model: the 3-level DCNet model
:param data: a tuple containing training and testing data, like &＃96;((x_train, y_train), (x_test, y_test))&＃96;
:param args: arguments
:return: The trained model
"""
# unpacking the data
(x_train, y_train), (x_test, y_test) &＃61; data
row &＃61; x_train.shape[1]
col &＃61; x_train.shape[2]
channel &＃61; x_train.shape[3]
# callbacks
log &＃61; callbacks.CSVLogger(args.save_dir &＃43; &＃39;/log.csv&＃39;)
tb &＃61; callbacks.TensorBoard(log_dir&＃61;args.save_dir &＃43; &＃39;/tensorboard-logs&＃39;, histogram_freq&＃61;int(args.debug))
checkpoint &＃61; callbacks.ModelCheckpoint(args.save_dir &＃43; &＃39;/weights-{epoch:02d}.h5&＃39;, monitor&＃61;&＃39;val_capsnet_acc&＃39;,
verbose&＃61;1)
lr_decay &＃61; callbacks.LearningRateScheduler(schedule&＃61;lambda epoch: args.lr * (args.lr_decay ** epoch))
# compile the model
# Notice the four separate losses (for separate backpropagations)
model.compile(optimizer&＃61;optimizers.Adam(lr&＃61;args.lr),
loss&＃61;[margin_loss, &＃39;mse&＃39;],
loss_weights&＃61;[1., args.lam_recon],
metrics&＃61;{&＃39;capsnet&＃39;: &＃39;accuracy&＃39;})
#model.load_weights(&＃39;result/weights.h5&＃39;)
"""
# Training without data augmentation:
model.fit([x_train, y_train], [y_train, x_train], batch_size&＃61;args.batch_size, epochs&＃61;args.epochs,
validation_data&＃61;[[x_test, y_test], [y_test, x_test]], callbacks&＃61;[log, tb, checkpoint, lr_decay])
"""
# Training with data augmentation
def train_generator(x, y, batch_size, shift_fraction&＃61;0.):
train_datagen &＃61; ImageDataGenerator(width_shift_range&＃61;shift_fraction,
height_shift_range&＃61;shift_fraction) # shift up to 2 pixel for MNIST
generator &＃61; train_datagen.flow(x, y, batch_size&＃61;batch_size)
while 1:
x_batch, y_batch &＃61; generator.next()
yield ([x_batch, y_batch], [y_batch, x_batch])
# Training with data augmentation. If shift_fraction&＃61;0., also no augmentation.
model.fit_generator(generator&＃61;train_generator(x_train, y_train, args.batch_size, args.shift_fraction),
steps_per_epoch&＃61;int(y_train.shape[0] / args.batch_size),
epochs&＃61;args.epochs,
validation_data&＃61;[[x_test, y_test], [y_test, x_test]],
callbacks&＃61;[log, tb, checkpoint, lr_decay])
# Save model weights
model.save_weights(args.save_dir &＃43; &＃39;/trained_model.h5&＃39;)
print(&＃39;Trained model saved to \&＃39;%s/trained_model.h5\&＃39;&＃39; % args.save_dir)
plot_log(args.save_dir &＃43; &＃39;/log.csv&＃39;, show&＃61;True)
return model
开发者ID:ssrp&＃xff0c;项目名称:Multi-level-DCNet&＃xff0c;代码行数:62&＃xff0c;
示例20: train
点赞 4
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def train(model, data, args):
"""
Training a CapsuleNet
:param model: the CapsuleNet model
:param data: a tuple containing training and testing data, like &＃96;((x_train, y_train), (x_test, y_test))&＃96;
:param args: arguments
:return: The trained model
"""
# unpacking the data
(x_train, y_train), (x_test, y_test) &＃61; data
# callbacks
log &＃61; callbacks.CSVLogger(args.save_dir &＃43; &＃39;/log.csv&＃39;)
tb &＃61; callbacks.TensorBoard(log_dir&＃61;args.save_dir &＃43; &＃39;/tensorboard-logs&＃39;,
batch_size&＃61;args.batch_size, histogram_freq&＃61;args.debug)
checkpoint &＃61; callbacks.ModelCheckpoint(args.save_dir &＃43; &＃39;/weights-{epoch:02d}.h5&＃39;, monitor&＃61;&＃39;val_capsnet_acc&＃39;,
save_best_only&＃61;True, save_weights_only&＃61;True, verbose&＃61;1)
lr_decay &＃61; callbacks.LearningRateScheduler(schedule&＃61;lambda epoch: args.lr * (0.95 ** epoch))
# compile the model
model.compile(optimizer&＃61;optimizers.Adam(lr&＃61;args.lr),
loss&＃61;[margin_loss, &＃39;mse&＃39;],
loss_weights&＃61;[1., args.lam_recon],
metrics&＃61;{&＃39;capsnet&＃39;: &＃39;accuracy&＃39;})
"""
# Training without data augmentation:
model.fit([x_train, y_train], [y_train, x_train], batch_size&＃61;args.batch_size, epochs&＃61;args.epochs,
validation_data&＃61;[[x_test, y_test], [y_test, x_test]], callbacks&＃61;[log, tb, checkpoint, lr_decay])
"""
# Begin: Training with data augmentation ---------------------------------------------------------------------#
def train_generator(x, y, batch_size, shift_fraction&＃61;0.):
train_datagen &＃61; ImageDataGenerator(width_shift_range&＃61;shift_fraction,
height_shift_range&＃61;shift_fraction,
horizontal_flip&＃61;True) # shift up to 2 pixel for MNIST
generator &＃61; train_datagen.flow(x, y, batch_size&＃61;batch_size)
while 1:
x_batch, y_batch &＃61; generator.next()
yield ([x_batch, y_batch], [y_batch, x_batch])
# Training with data augmentation. If shift_fraction&＃61;0., also no augmentation.
model.fit_generator(generator&＃61;train_generator(x_train, y_train, args.batch_size, args.shift_fraction),
steps_per_epoch&＃61;int(y_train.shape[0] / args.batch_size),
epochs&＃61;args.epochs,
validation_data&＃61;[[x_test, y_test], [y_test, x_test]],
callbacks&＃61;[log, tb, checkpoint, lr_decay])
# End: Training with data augmentation -----------------------------------------------------------------------#
model.save_weights(args.save_dir &＃43; &＃39;/trained_model.h5&＃39;)
print(&＃39;Trained model saved to \&＃39;%s/trained_model.h5\&＃39;&＃39; % args.save_dir)
from utils import plot_log
plot_log(args.save_dir &＃43; &＃39;/log.csv&＃39;, show&＃61;True)
return model
开发者ID:XifengGuo&＃xff0c;项目名称:CapsNet-Fashion-MNIST&＃xff0c;代码行数:58&＃xff0c;
示例21: test_stop_training_csv
点赞 4
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def test_stop_training_csv(tmpdir):
np.random.seed(1337)
fp &＃61; str(tmpdir / &＃39;test.csv&＃39;)
(X_train, y_train), (X_test, y_test) &＃61; get_test_data(num_train&＃61;train_samples,
num_test&＃61;test_samples,
input_shape&＃61;(input_dim,),
classification&＃61;True,
num_classes&＃61;num_classes)
y_test &＃61; np_utils.to_categorical(y_test)
y_train &＃61; np_utils.to_categorical(y_train)
cbks &＃61; [callbacks.TerminateOnNaN(), callbacks.CSVLogger(fp)]
model &＃61; Sequential()
for _ in range(5):
model.add(Dense(num_hidden, input_dim&＃61;input_dim, activation&＃61;&＃39;relu&＃39;))
model.add(Dense(num_classes, activation&＃61;&＃39;linear&＃39;))
model.compile(loss&＃61;&＃39;mean_squared_error&＃39;,
optimizer&＃61;&＃39;rmsprop&＃39;)
def data_generator():
i &＃61; 0
max_batch_index &＃61; len(X_train) // batch_size
tot &＃61; 0
while 1:
if tot > 3 * len(X_train):
yield np.ones([batch_size, input_dim]) * np.nan, np.ones([batch_size, num_classes]) * np.nan
else:
yield (X_train[i * batch_size: (i &＃43; 1) * batch_size],
y_train[i * batch_size: (i &＃43; 1) * batch_size])
i &＃43;&＃61; 1
tot &＃43;&＃61; 1
i &＃61; i % max_batch_index
history &＃61; model.fit_generator(data_generator(),
len(X_train) // batch_size,
validation_data&＃61;(X_test, y_test),
callbacks&＃61;cbks,
epochs&＃61;20)
loss &＃61; history.history[&＃39;loss&＃39;]
assert len(loss) > 1
assert loss[-1] &＃61;&＃61; np.inf or np.isnan(loss[-1])
values &＃61; []
with open(fp) as f:
for x in reader(f):
values.append(x)
assert &＃39;nan&＃39; in values[-1], &＃39;The last epoch was not logged.&＃39;
os.remove(fp)
开发者ID:hello-sea&＃xff0c;项目名称:DeepLearning_Wavelet-LSTM&＃xff0c;代码行数:51&＃xff0c;
示例22: test_CSVLogger
点赞 4
# 需要导入模块: from keras import callbacks [as 别名]
# 或者: from keras.callbacks import CSVLogger [as 别名]
def test_CSVLogger(tmpdir):
np.random.seed(1337)
filepath &＃61; str(tmpdir / &＃39;log.tsv&＃39;)
sep &＃61; &＃39;\t&＃39;
(X_train, y_train), (X_test, y_test) &＃61; get_test_data(num_train&＃61;train_samples,
num_test&＃61;test_samples,
input_shape&＃61;(input_dim,),
classification&＃61;True,
num_classes&＃61;num_classes)
y_test &＃61; np_utils.to_categorical(y_test)
y_train &＃61; np_utils.to_categorical(y_train)
def make_model():
np.random.seed(1337)
model &＃61; Sequential()
model.add(Dense(num_hidden, input_dim&＃61;input_dim, activation&＃61;&＃39;relu&＃39;))
model.add(Dense(num_classes, activation&＃61;&＃39;softmax&＃39;))
model.compile(loss&＃61;&＃39;categorical_crossentropy&＃39;,
optimizer&＃61;optimizers.SGD(lr&＃61;0.1),
metrics&＃61;[&＃39;accuracy&＃39;])
return model
# case 1, create new file with defined separator
model &＃61; make_model()
cbks &＃61; [callbacks.CSVLogger(filepath, separator&＃61;sep)]
model.fit(X_train, y_train, batch_size&＃61;batch_size,
validation_data&＃61;(X_test, y_test), callbacks&＃61;cbks, epochs&＃61;1)
assert os.path.isfile(filepath)
with open(filepath) as csvfile:
dialect &＃61; Sniffer().sniff(csvfile.read())
assert dialect.delimiter &＃61;&＃61; sep
del model
del cbks
# case 2, append data to existing file, skip header
model &＃61; make_model()
cbks &＃61; [callbacks.CSVLogger(filepath, separator&＃61;sep, append&＃61;True)]
model.fit(X_train, y_train, batch_size&＃61;batch_size,
validation_data&＃61;(X_test, y_test), callbacks&＃61;cbks, epochs&＃61;1)
# case 3, reuse of CSVLogger object
model.fit(X_train, y_train, batch_size&＃61;batch_size,
validation_data&＃61;(X_test, y_test), callbacks&＃61;cbks, epochs&＃61;1)
import re
with open(filepath) as csvfile:
output &＃61; " ".join(csvfile.readlines())
assert len(re.findall(&＃39;epoch&＃39;, output)) &＃61;&＃61; 1
os.remove(filepath)
assert not tmpdir.listdir()
开发者ID:hello-sea&＃xff0c;项目名称:DeepLearning_Wavelet-LSTM&＃xff0c;代码行数:55&＃xff0c;
注&＃xff1a;本文中的keras.callbacks.CSVLogger方法示例整理自Github/MSDocs等源码及文档管理平台&＃xff0c;相关代码片段筛选自各路编程大神贡献的开源项目&＃xff0c;源码版权归原作者所有&＃xff0c;传播和使用请参考对应项目的License&＃xff1b;未经允许&＃xff0c;请勿转载。
京公网安备 11010802041100号 | 京ICP备19059560号-4 | PHP1.CN 第一PHP社区 版权所有