#Author: Michail Mamalakis #Version: 0.1 #Licence: #email:[email protected] from __future__ import division, print_function from covid_pipeline import main # name of weights store of main segmentation mn = main.main('covid_80') # run train of main segmentation mn.train_run() # run test of main segmentation # mn.test_run()
#Author: Michail Mamalakis #Version: 0.1 #Licence: #email:[email protected] from __future__ import division, print_function from covid_pipeline import main # name of weights store of main segmentation mn = main.main('sgd') # run train of main segmentation mn.train_run() # run test of main segmentation # mn.test_run()
#Author: Michail Mamalakis #Version: 0.1 #Licence: #email:[email protected] from __future__ import division, print_function from covid_pipeline import main # name of weights store of main segmentation mn = main.main('covid_300') # run train of main segmentation mn.train_run() # run test of main segmentation # mn.test_run()
def run_training(self, loss_type, model_structure, X, Y):
# MPU: data paraller mode deep learning
# GPU: model paraller mode deep learning
#one-hot encoder mask binary labels
#if self.path_case=='main':
# Y = utils.to_categorical(Y,(np.array(Y).shape[1]+1))
# Y=Y.reshape(np.array(Y).shape[0],np.array(Y).shape[2],np.array(Y).shape[3],np.array(Y).shape[4])
# print(np.array(Y).shape)
if self.ram == 'GPU':
model_structure = multi_gpu_model(model_structure, gpus=self.ngpu)
self.batch_size = self.batch_size * self.ngpu
#define metrics for training
metrics_algorithm = self.metrics_generator()
loss = self.load_loss(loss_type)
model_structure.compile(optimizer=self.optimizer,
loss=loss,
metrics=metrics_algorithm)
if self.normalize_image == 'True':
X = regularization.normalize(X)
if self.shuffle == 'True':
# shuffle images and masks in parallel
rng_state = np.random.get_state()
np.random.shuffle(X)
np.random.set_state(rng_state)
np.random.shuffle(Y)
#define callbacks
self.model_json = model_structure.to_json()
self.callbacks = self.callbacks_define(self.monitor, self.weight_name)
#validation data
cn = create_net.create_net(self.main_model)
if self.validation == 'on':
for cross_val_num in range(int(self.cross_validation_number)):
print("cross validation run: ", cross_val_num, "/",
int(self.cross_validation_number))
if cross_val_num != 0:
model_structure = cn.net([], [], self.path_case,
self.height, self.channels,
(self.classes), self.width)
rng_state = np.random.get_state()
np.random.shuffle(X)
np.random.set_state(rng_state)
np.random.shuffle(Y)
model_structure.compile(optimizer=self.optimizer,
loss=loss,
metrics=metrics_algorithm)
self.callbacks = self.callbacks_define(
self.monitor,
(self.weight_name + '_' + str(cross_val_num)))
h_d = handle_data.handle_data(X, Y, self.path_case)
training_augment, train_steps_per_epoch, validation_augment, val_steps_per_epoch = h_d.validation_data(
)
if self.ram == 'CPU':
history = model_structure.fit_generator(
training_augment,
epochs=self.epochs,
steps_per_epoch=train_steps_per_epoch,
verbose=1,
callbacks=self.callbacks,
validation_data=validation_augment,
validation_steps=val_steps_per_epoch)
if self.ram == 'MPU':
self.batch_size = self.batch_size
history = model_structure.fit_generator(
training_augment,
epochs=self.epochs,
steps_per_epoch=train_steps_per_epoch,
verbose=1,
callbacks=self.callbacks,
validation_data=validation_augment,
validation_steps=val_steps_per_epoch,
workers=self.num_cores,
use_multiprocessing=True)
#call test for evaluate results..P.S. need modification this part: TODDO!!!!
mn = main.main(self.path_case)
print(h_d.Xval.shape, h_d.Yval.shape)
mn.X = np.array(h_d.Xval)
mn.Y = np.array(h_d.Yval)
print('test output')
print(mn.X[1, :, :, 1], mn.Y[1, :])
filepath = str(self.weight_name + '_' + str(cross_val_num))
file = str(self.weight_name)
if cross_val_num != 0:
mn.test_run(
model_structure, [filepath],
'/fastdata/mer17mm/private/Data/',
['tuberculosis', 'pneumonia', 'healthy', 'COVID-19'],
2)
else:
mn.test_run(
model_structure, [file],
'/fastdata/mer17mm/private/Data/',
['tuberculosis', 'pneumonia', 'healthy', 'COVID-19'],
2)
else:
h_d2 = handle_data.handle_data(X, Y, self.path_case)
training_augment, train_steps_per_epoch = h_d2.no_validation_data()
if self.ram == 'CPU':
history = model_structure.fit_generator(
training_augment,
epochs=self.epochs,
steps_per_epoch=train_steps_per_epoch,
verbose=1,
callbacks=self.callbacks)
if self.ram == 'MPU':
self.batch_size = self.batch_size
history = model_structure.fit_generator(
training_augment,
epochs=self.epochs,
steps_per_epoch=train_steps_per_epoch,
verbose=1,
callbacks=self.callbacks,
workers=self.num_cores,
use_multiprocessing=True)
mn = main.main(self.path_case)
mn.Y = np.array(h_d2.Y)
mn.X = np.array(h_d2.X)
print(h_d2.X.shape, h_d2.Y.shape)
mn.test_run(model_structure, [file],
'/fastdata/mer17mm/private/Data/',
['tuberculosis', 'COVID-19', 'healthy', 'pneumonia'],
2)
return model_structure, history
#Author: Michail Mamalakis #Version: 0.1 #Licence: #email:[email protected] from __future__ import division, print_function from covid_pipeline import main # name of weights store of main segmentation mn = main.main('unet') # run train of main segmentation mn.train_run() # run test of main segmentation # mn.test_run()