def main():
direc_data = '/data/npz_data/cells/unspecified_nuclear_data/nuclear_movie/'
dataset = 'nuclear_movie_same'
training_data = np.load('{}{}.npz'.format(direc_data, dataset))
optimizer = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
lr_sched = rate_scheduler(lr=0.01, decay=0.99)
in_shape = (14, 14, 1)
model = the_model(input_shape=in_shape) #, n_features=1, reg=1e-5)
train_model_siamese(
model=model,
dataset='nuclear_movie_same',
optimizer=optimizer,
expt='',
it=0,
batch_size=1,
n_epoch=100,
direc_save='/data/models/cells/unspecified_nuclear_data/nuclear_movie',
direc_data=
'/data/npz_data/cells/unspecified_nuclear_data/nuclear_movie/',
lr_sched=lr_sched,
rotation_range=0,
flip=True,
shear=0,
class_weight=None)
def train_model_on_training_data():
direc_save = os.path.join(MODEL_DIR, PREFIX)
direc_data = os.path.join(NPZ_DIR, PREFIX)
training_data = np.load(os.path.join(direc_data, DATA_FILE + '.npz'))
class_weights = training_data['class_weights']
X, y = training_data['X'], training_data['y']
print('X.shape: {}\ny.shape: {}'.format(X.shape, y.shape))
n_epoch = 100
batch_size = 32 if DATA_OUTPUT_MODE == 'sample' else 1
optimizer = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
lr_sched = rate_scheduler(lr=0.01, decay=0.99)
model_args = {'norm_method': 'median', 'reg': 1e-5, 'n_features': 3}
data_format = K.image_data_format()
row_axis = 2 if data_format == 'channels_first' else 1
col_axis = 3 if data_format == 'channels_first' else 2
channel_axis = 1 if data_format == 'channels_first' else 3
if DATA_OUTPUT_MODE == 'sample':
train_model = train_model_sample
the_model = bn_feature_net_61x61
model_args['n_channels'] = 1
elif DATA_OUTPUT_MODE == 'conv' or DATA_OUTPUT_MODE == 'disc':
train_model = train_model_conv
the_model = bn_dense_feature_net
model_args['location'] = False
size = (RESHAPE_SIZE,
RESHAPE_SIZE) if RESIZE else X.shape[row_axis:col_axis + 1]
if data_format == 'channels_first':
model_args['input_shape'] = (X.shape[channel_axis], size[0],
size[1])
else:
model_args['input_shape'] = (size[0], size[1],
X.shape[channel_axis])
model = the_model(**model_args)
train_model(model=model,
dataset=DATA_FILE,
optimizer=optimizer,
batch_size=batch_size,
n_epoch=n_epoch,
direc_save=direc_save,
direc_data=direc_data,
lr_sched=lr_sched,
class_weight=class_weights,
rotation_range=180,
flip=True,
shear=True)
import numpy as np
batch_size = 1
n_epoch = 100
dataset = "nuclei_broad_same_conv_61x61"
expt = "bn_dense_feature_net"
direc_save = "/data/trained_networks/nuclei_broad/"
direc_data = "/data/training_data_npz/nuclei_broad/"
# Create output ditrectory, if necessary
pathlib.Path(direc_save).mkdir(parents=True, exist_ok=True)
optimizer = SGD(lr=0.01, decay=1e-6, momentum=0.9, nesterov=True)
lr_sched = rate_scheduler(lr=0.01, decay=0.99)
file_name = os.path.join(direc_data, dataset + ".npz")
training_data = np.load(file_name)
class_weights = training_data["class_weights"]
for iterate in range(1):
model = the_model(batch_shape=(1, 512, 512, 1),
n_features=3,
reg=1e-5,
softmax=True,
permute=True)
train_model(model=model,
dataset=dataset,
from scipy.misc import imsave
batch_size = 1
n_epoch = 50
dataset = "nuclear_movie_disc_same"
expt = "bn_dense_net_3D"
direc_save = "/data/trained_networks/nuclear_movie/"
direc_data = "/data/training_data_npz/nuclear_movie/"
# Create output ditrectory, if necessary
pathlib.Path(direc_save).mkdir(parents=True, exist_ok=True)
optimizer = SGD(lr=1e-2, decay=1e-6, momentum=0.9, nesterov=True)
lr_sched = rate_scheduler(lr=1e-2, decay=0.99)
file_name = os.path.join(direc_data, dataset + ".npz")
training_data = np.load(file_name)
for iterate in range(1):
model = the_model(batch_shape=(1, 1, 5, 256, 256),
n_features=3,
reg=1e-5,
location=False,
permute=True,
softmax=False)
trained_model = train_model(model=model,
dataset=dataset,