def min_len():
Path = ['nnet_data_logpower_noise_chans_train_normalized/dt05_caf_real','nnet_data_logpower_noise_chans_train_normalized/dt05_bus_real',
'nnet_data_logpower_noise_chans_train_normalized/dt05_ped_real', 'nnet_data_logpower_noise_chans_train_normalized/dt05_str_real']
# ,'nnet_data_logpower_1/dt05_caf_real','nnet_data_logpower_1/dt05_ped_real','nnet_data_logpower_1/dt05_str_real'
#,'nnet_data_logpower_noise/dt05_str_real','nnet_data_logpower_noise/dt05_ped_real'
# 'chans_noadapt_ds_real/dt05_bus_real','chans_noadapt_ds_real/dt05_caf_real','chans_noadapt_ds_real/dt05_ped_real','chans_noadapt_ds_real/dt05_str_real'
print("min_len")
Filenames = np.array([])
for p in Path:
for root, dirs, files in os.walk(p):
for filename in files:
if filename.split('.')[1] == 'phase' or filename.split('.')[1] == 'predicted':
continue
fp = os.path.join(p, filename)
Filenames = np.append(Filenames,fp)
min_len = 100000000
for fp in Filenames:
real_wiener_all, wiener_all = readmat(fp, read='inout', keys=('Input', 'Output'))
if len(wiener_all)<min_len:
min_len = len(wiener_all)
return min_len
def make_ds(Filenames,context,zero_delimiter = False):
X,Y = 0,0
if zero_delimiter:
xdel = np.zeros((1, context, 257))
ydel = np.zeros((1, context, 257 * 5))
for fp in Filenames:
real_wiener_all, wiener_all = readmat(fp, read='inout', keys=('Input', 'Output'))
dslength = len(wiener_all)
X_ = np.zeros((dslength, 1, 257 * 5))
Y_ = np.zeros((dslength, 1, 257))
for i in range(np.size(X_, 0)):
x = wiener_all[i].reshape((1, 257 * 5))
# x = np.array([max(x[0][i],-35) for i in range (1285)])
X_[i][0] = x
for i in range(np.size(Y_, 0)):
y = real_wiener_all[i].reshape((1, 257))
# y = np.array([max(y[0][i], -35) for i in range(257)])
Y_[i][0] = y
if context:
X_,Y_ = create_dataset(X_,Y_,context)
X_ = np.squeeze(X_)
Y_ = np.squeeze(Y_)
if zero_delimiter:
newX_ = np.zeros((len(X_)+1,context,257*5))
newY_ = np.zeros((len(Y_)+1,context,257))
newX_[1:len(X_)+1,:,:] = X_
newY_[1:len(Y_)+1,:,:] = Y_
X_=newX_
Y_ =newY_
if type(X) != type(X_):
X = X_
Y = Y_
else:
X = np.concatenate((X, X_))
Y = np.concatenate((Y, Y_))
print(X.shape)
return X,Y
def main():
real_wiener_all, wiener_all = readmat(
'nnet_data_logpower_noise/dt05_bus_real/F01_050C010K_BUS.io.mat',
keys=('Input', 'Output'),
read='inout')
X = np.zeros((len(wiener_all), 1, 257 * 5))
for i in range(np.size(X, 0)):
x = wiener_all[i].reshape((1, 257 * 5))
X[i][0] = x
Y = np.zeros((len(wiener_all), 1, 257))
for i in range(np.size(Y, 0)):
y = real_wiener_all[i].reshape((1, 257))
Y[i][0] = y
X1, Y1 = create_dataset(X, Y, 400)
print(X1.shape, Y1.shape)
print(X.shape, Y.shape)
from keras.models import Sequential
from keras.layers import Dense
from keras.models import model_from_yaml
from keras import losses
from read_data import readmat
from numpy import array,log,exp
import numpy as np
from theano import compile
import matplotlib.pyplot as plt
dslength = 390
import numpy
import os
compile.mode.Mode(linker='py', optimizer=None)
real_wiener_all,wiener_all= array(readmat('network_input.mat'))
#wiener_all_cv, real_wiener_all_cv = array(readmat('network_cv.mat'))
X = np.zeros((dslength,257))
Y = np.zeros((dslength,257))
wiener_all = array(wiener_all)
real_wiener_all = array(real_wiener_all)
for i in range( dslength):
for j in range(257):
X1 = wiener_all[i][j]
#X1 = [X1[k] + 0.01 for k in range(len(X1))]
X[i][j] = X1
Y1 = real_wiener_all[i][j]
# Y1 = [Y1[k] + 0.01 for k in range(len(Y1))]
Y[i][j] = Y1
yaml_file = open(yaml_file_name, 'r')
loaded_model_yaml = yaml_file.read()
yaml_file.close()
loaded_model = model_from_yaml(loaded_model_yaml)
# load weights into new model
loaded_model.load_weights(h5_file_name)
print("Loaded model from disk")
loaded_model.compile(loss=losses.mean_squared_logarithmic_error,
optimizer='rmsprop',
metrics=['mse', 'mae', 'logcosh'])
for filename in Filenames:
wiener_all = readmat(filename, keys=('Input'), read='in')
X = np.zeros((len(wiener_all), 1, 257 * 5))
for i in range(np.size(X, 0)):
x = wiener_all[i].reshape((1, 257 * 5))
X[i][0] = x
"""
Y = np.zeros((len(wiener_all),1,257))
for i in range(np.size(Y, 0)):
y = real_wiener_all[i].reshape((1, 257))
Y[i][0] = y
"""
# X, Y = create_dataset(X, Y, 2)
def main():
best = [
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_051C0104_PED.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_050C010N_PED.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_053C0111_PED.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_22HC010R_PED.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_052C010J_PED.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_421C020Q_PED.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_22GC010D_PED.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_420C020P_PED.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_ped_real/M03_422C020F_PED.io.mat'
]
worst = [
'nnet_data_logpower_noise_chans_train/dt05_bus_real/F01_422C0213_BUS.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_bus_real/F01_420C020K_BUS.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_bus_real/F01_051C0104_BUS.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_str_real/F01_053C0115_STR.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_str_real/F04_423C0204_STR.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_bus_real/F01_051C0111_BUS.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_str_real/F04_423C020I_STR.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_str_real/F01_421C020A_STR.io.mat',
'nnet_data_logpower_noise_chans_train/dt05_str_real/F01_22GC0113_STR.io.mat'
]
yaml_file_name = 'lstm_5ch_2_layers_real_chans.yaml'
h5_file_name = 'lstm_5ch_2_layers_real_chans.h5'
yaml_file = open(yaml_file_name, 'r')
loaded_model_yaml = yaml_file.read()
yaml_file.close()
loaded_model = model_from_yaml(loaded_model_yaml)
# load weights into new model
loaded_model.load_weights(h5_file_name)
print("Loaded model from disk")
loaded_model.compile(loss=losses.mean_squared_logarithmic_error,
optimizer='rmsprop',
metrics=['mse', 'mae', 'logcosh'])
for filename in worst:
try:
real_wiener_all, wiener_all = readmat(filename,
keys=('Input', 'Output'),
read='inout')
print(len(real_wiener_all))
except:
print(filename, 'error')
continue
X = np.zeros((len(wiener_all), 1, 257 * 5))
for i in range(np.size(X, 0)):
x = wiener_all[i].reshape((1, 257 * 5))
X[i][0] = x
Y = np.zeros((len(wiener_all), 1, 257))
for i in range(np.size(Y, 0)):
y = real_wiener_all[i].reshape((1, 257))
Y[i][0] = y
Y1 = loaded_model.predict(X, verbose=1)
num = -1
bestloss = 0
for i in range(len(Y1)):
loss = mean_squared_logarithmic_error(Y[i][0][0:60],
Y1[i][0][0:60])
if bestloss < loss:
bestloss = loss
num = i
#Sum_loss +=mean_squared_logarithmic_error(Y[i][0][0:60],Y1[i][0][0:60])
#Sum_loss +=losses.mean_absolute_error(Y[i][0][0:60],Y1[i][0][0:60]).eval()
Ys = Y[i][0]
Ys = np.expand_dims(Ys, 2)
Y1s = Y1[i][0]
Y1s = np.expand_dims(Y1s, 2)
I = X[i][0]
pre = plt.plot(Y1s[0:257], label='predicted')
inp = plt.plot(Ys[0:257], label='Output')
#plt.legend()
#plt.show()
print(filename, num, bestloss)
for i in range(np.size(Y, 0)):
y = real_wiener_all[i].reshape((1, 257))
Y[i][0] = y
Y1 = loaded_model.predict(X,verbose = 1)
losses_stat[filename] = np.mean(losses.mean_squared_logarithmic_error(Y,Y1).eval())
"""
noise_type_losses = {'bus': [], 'caf': [], 'str': [], 'ped': []}
for noise_type in namelist.keys():
for filename in namelist[noise_type]:
try:
real_wiener_all, wiener_all = readmat(filename,
keys=('Input', 'Output'),
read='inout')
except:
print(filename, 'error')
continue
X = np.zeros((len(wiener_all), 1, 257 * 5))
for i in range(np.size(X, 0)):
x = wiener_all[i].reshape((1, 257 * 5))
X[i][0] = x
Y = np.zeros((len(wiener_all), 1, 257))
for i in range(np.size(Y, 0)):
y = real_wiener_all[i].reshape((1, 257))