print('Compute node: {}'.format(compute_node))
else:
os.environ["CUDA_VISIBLE_DEVICES"]= "-1"
# Turn on soft memory allocation
tf_config = tf.compat.v1.ConfigProto()
tf_config.gpu_options.allow_growth = True
tf_config.log_device_placement = False
sess = tf.compat.v1.Session(config=tf_config)
#K.v1.set_session(sess)
# In[4]:
data_loader = data_selector(cgf['DATASET']['name'], cgf['DATASET']['arguments'])
data, labels, diff = data_loader.load_data()
# In[5]:
data = data[:3000,:,:]
labels = labels[:3000]
'''
data = np.load(cgf["DATASET"]["arguments"]["images"])
labels = np.load(cgf["DATASET"]["arguments"]["labels"])
'''
import yaml
import sys
from data_bank import data_selector
if __name__ == '__main__':
configfile = 'config_files/create_images_shades.yml'
with open(configfile) as ymlfile:
cgf = yaml.load(ymlfile, Loader=yaml.SafeLoader)
data_loader = data_selector(cgf['DATA']['name'], cgf['DATA']['arguments'])
images, labels, diff = data_loader.load_data()
print('Data done.')
configfile = 'config_files/config_visualise.yml'
with open(configfile) as ymlfile:
cgf = yaml.load(ymlfile, Loader=yaml.SafeLoader)
configfile1 = 'config_files/config.yml'
with open(configfile1) as ymlfile:
cgf1 = yaml.load(ymlfile, Loader=yaml.SafeLoader)
model_id = cgf['data_arguments']['model']
model_dest = cgf['data_arguments']['model_dest']
model_path = join(model_dest, str(model_id))
data_name = cgf['data_name']
data_arguments = cgf['data_arguments']
data_loader = data_selector(data_name, data_arguments)
print(data_loader)
data_images, data_labels, data_diff = data_loader.load_data()
model_name = cgf1['MODEL']['name']
model_arguments = cgf1['MODEL']['arguments']
input_shape = data_images.shape[1:]
output_shape = data_labels.shape[1]
# Set the default precision
model_precision = cgf1['MODEL_METADATA']['precision']
K.set_floatx(model_precision)
model = mb.model_selector(model_name, input_shape, output_shape,
model_arguments)
import numpy as np
import yaml
from data_bank import data_selector
import os
from os.path import join
import sys
from shade2_algo import shades2_alg
if __name__ == "__main__":
configfile = 'config_files/config_exist.yml'
with open(configfile) as ymlfile:
cgf = yaml.load(ymlfile, Loader=yaml.SafeLoader)
data_loader_test = data_selector(cgf['DATASET_TEST']['name'], cgf['DATASET_TEST']['arguments'])
test_data, test_labels, _ = data_loader_test.load_data()
score = 0
for i in range(len(test_labels)):
if shades2_alg(test_data[i])==test_labels[i]:
score += 1
print('Accuracy on test set: {}%'.format(100*score/len(test_labels)))
os.environ['CUDA_VISIBLE_DEVICES']= "-1"
with open('config_files/config_evaluate.yml') as ymlfile:
cgf_eval = yaml.load(ymlfile, Loader= yaml.SafeLoader)
model_id = cgf_eval['MODEL']['model_id']
model_dest = cgf_eval['MODEL']['model_dest']
model_path = join(model_dest, str(model_id))
with open(join(model_path, 'config.yml')) as ymlfile:
cgf = yaml.load(ymlfile, Loader= yaml.SafeLoader)
# Select dataset
use_default_test = cgf_eval['DATASET']['use_default']
if use_default_test:
data_loader = data_selector(cgf['DATASET_TEST']['name'], cgf['DATASET_TEST']['arguments'])
else:
data_loader = data_selector(cgf_eval['DATASET']['name'], cgf_eval['DATASET']['arguments'])
print('\nDATASET TEST')
print(data_loader)
test_data, test_labels, test_diff = data_loader.load_data()
print(test_data.shape)
model_name = cgf['MODEL']['name']
model_arguments = cgf['MODEL']['arguments']
input_shape = test_data.shape[1:]
output_shape = test_labels.shape[1];
if use_gpu:
compute_node = cgf['COMPUTER_SETUP']['compute_node']
os.environ["CUDA_VISIBLE_DEVICES"] = "%d" % (compute_node)
print('Compute node: {}'.format(compute_node))
else:
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
# Turn on soft memory allocation
tf_config = tf.compat.v1.ConfigProto()
tf_config.gpu_options.allow_growth = True
tf_config.log_device_placement = False
sess = tf.compat.v1.Session(config=tf_config)
#K.v1.set_session(sess)
# Load train and validatation data
data_loader_train = data_selector(cgf['DATASET_TRAIN']['name'],
cgf['DATASET_TRAIN']['arguments'])
data_loader_validate = data_selector(cgf['DATASET_VAL']['name'],
cgf['DATASET_VAL']['arguments'])
print('\nDATASET TRAIN')
print(data_loader_train)
print('DATASET VALIDATION')
print(data_loader_validate)
train_data, train_labels, _ = data_loader_train.load_data()
val_data, val_labels, _ = data_loader_validate.load_data()
# Trying for resnet since memory seems to be the issue
train_data = train_data[:3000, :, :]
train_labels = train_labels[:3000]
if (cgf['MODEL']['name'] == "resnet"):