def DATA(train_with, test_with, data_path, mode):
if train_with == 'clean':
trX, trY = load_specific_noisy_data(data_path,
is_training='TRAIN',
mode=mode,
noise_name='clean')
elif train_with == 'noisy':
trX, trY = load_random_noisy_data(data_path, 'TRAIN', mode)
else:
raise ValueError('should set "--train_with" value')
if test_with == 'clean':
vaX, vaY = load_specific_noisy_data(data_path,
is_training='VALID',
mode=mode,
noise_name='clean')
teX, teY = load_specific_noisy_data(data_path,
is_training='TEST',
mode=mode,
noise_name='clean')
elif test_with == 'noisy':
vaX, vaY = load_random_noisy_data(data_path, 'VALID', mode)
teX, teY = load_random_noisy_data(data_path, 'TEST', mode)
else:
raise ValueError('should set "--test_with" value')
return trX, trY, vaX, vaY, teX, teY #shape [sample, 120(f), 99?(t)], [sample,]
def DATA(is_training, train_with, test_with, data_path, mode):
if is_training == 'TRAIN':
if train_with == 'clean':
trX, trY = load_specific_noisy_data(data_path,
is_training='TRAIN',
mode=mode,
noise_name='clean')
elif train_with == 'noisy':
trX1, trY1 = load_specific_noisy_data(data_path,
is_training='TRAIN',
mode=mode,
noise_name='clean')
print(str(trX1.shape), str(trY1.shape))
trX2, trY2 = load_specific_noisy_data(data_path,
is_training='TRAIN',
mode=mode,
noise_name='white_noise')
print(str(trX2.shape), str(trY2.shape))
trX3, trY3 = load_specific_noisy_data(data_path,
is_training='TRAIN',
mode=mode,
noise_name='pink_noise')
print(str(trX3.shape), str(trY3.shape))
trX = np.concatenate((trX1, trX2, trX3), axis=0)
trY = np.concatenate((trY1, trY2, trY3), axis=0)
else:
raise ValueError('should set "--train_with" value')
if test_with == 'clean':
vaX, vaY = load_specific_noisy_data(data_path,
is_training='VALID',
mode=mode,
noise_name='clean')
elif test_with == 'noisy':
vaX, vaY = load_random_noisy_data(data_path, 'VALID', mode)
else:
raise ValueError('should set "--test_with" value')
trX = np.expand_dims(trX[:, :, :, 1], axis=3)
vaX = np.expand_dims(vaX[:, :, :, 1], axis=3)
print(str(trX.shape), str(trY.shape), str(vaX.shape),
str(vaY.shape))
return (trX, trY, vaX, vaY)
elif is_training == 'TEST':
teX, teY = load_specific_noisy_data(data_path,
is_training='TEST',
mode=mode,
noise_name='clean')
teX = np.expand_dims(teX[:, :, :, 1], axis=3)
print(str(teX.shape), str(teY.shape))
return (teX, teY) #shape [sample, 120(f), 99?(t)], [sample,]
raise ValueError('No weights are provided.')
else:
test_result = save_path + '/test_result.csv'
fd_test_result = open(test_result,'a')
fd_test_result.write('test on epoch '+str(args.keep)+'SNR'+str(args.SNR)+' dimension:'+str(args.dimension)+'\n')
fd_test_result.write('test_mode,label30_acc,label21_acc\n')
# clean test
print('*'*30 + 'clean exp' + '*'*30)
label30_acc, label21_acc = test(multi_model, data=data,args=args)
fd_test_result.write('clean,'+str(label30_acc)+','+str(label21_acc)+'\n')
fd_test_result.flush()
for i in range(6):
print('*'*30 + 'Noisy '+ str(i+2) +' exp' + '*'*30)
if args.test_by=='noise':teX, teY = load_specific_noisy_data(args.data_path, 'TEST', args.mode, args.feature_len, noise_list[i]);cprint(noise_list[i],'red')
elif args.test_by=='echo':teX, teY = load_specific_noisy_data(args.data_path, 'TEST', args.mode, args.feature_len, 'echo');cprint('echo','red')
else: teX, teY = load_random_noisy_data(args.data_path,'TEST',args.mode, args.feature_len, SNR=args.SNR)
teX = Dimension(teX,args.dimension)#teX = np.expand_dims(teX[:,:,:,1],axis=3)
data = (teX, teY)
label30_acc, label21_acc = test(multi_model, data=data,args=args)
fd_test_result.write('noisy'+str(i)+','+str(label30_acc)+','+str(label21_acc)+'\n')
fd_test_result.flush()
fd_test_result.close()
else:
raise ValueError('Wrong is_training value')#'could not find %c in %s' % (ch,str))
# Code end
# For not decreasing issue: https://github.com/XifengGuo/CapsNet-Keras/issues/48
else:
test_result = save_path + '/test_result.csv'
fd_test_result = open(test_result, 'a')
fd_test_result.write('test on epoch ' + str(args.keep) + 'SNR' +
str(args.SNR) + '\n')
fd_test_result.write('test_mode,label30_acc,label21_acc\n')
# clean test
print('*' * 30 + 'clean exp' + '*' * 30)
label30_acc, label21_acc = test(multi_model, data=data, args=args)
fd_test_result.write('clean,' + str(label30_acc) + ',' +
str(label21_acc) + '\n')
fd_test_result.flush()
for i in range(5):
print('*' * 30 + 'Noisy ' + str(i + 2) + ' exp' + '*' * 30)
teX, teY = load_random_noisy_data(args.data_path,
'TEST',
args.mode,
SNR=args.SNR)
teX = np.expand_dims(teX[:, :, :, 1], axis=3)
data = (teX, teY)
label30_acc, label21_acc = test(multi_model,
data=data,
args=args)
fd_test_result.write('noisy' + str(i) + ',' +
str(label30_acc) + ',' +
str(label21_acc) + '\n')
fd_test_result.flush()
fd_test_result.close()
else:
raise ValueError(
'Wrong is_training value') #'could not find %c in %s' % (ch,str))
# Code end
teX = du.Dimension(teX, args.dimension)
acc = test(multi_model,
data=(teX, teY),
args=args,
matrix_name='/ConfusionMatrix_' +
args.ex_name + '_noisy' + str(snr),
label_names=label_names)
# csv write
fd_test_result.write('noisy' + str(snr) + ',' +
str(acc) + '\n')
cprint('Test with Clean + SNR' + str(snr), 'red')
teX, teY = du.load_random_noisy_data(
args.data_path,
'TEST',
args.mode,
args.feature_len,
SNR=snr,
open_set=args.open_set,
open_labels=open_labels)
#teX = np.expand_dims(teX[:,:,:,1],axis=3)
teX = du.Dimension(teX, args.dimension)
acc = test(multi_model,
data=(teX, teY),
args=args,
matrix_name='/ConfusionMatrix_' +
args.ex_name + '_mixed' + str(snr),
label_names=label_names)
# csv write
fd_test_result.write('mixed' + str(snr) + ',' +
str(acc) + '\n')
except: