def test_trainSignal(self):
maxFeatures = 10
# sig, force, fs, firings = getSignal(1, 30, "./signals/")
trainSigs = []
trainFirings = []
for i in range(10, 17):
sig, force, fs, firings = getSignal2(
"002041436" + str(i),
filePath="../signals/Hamid/00MaHaLI1002041436/")
trainSigs.append(sig)
trainFirings.append(firings)
sig2, force2, fs2, firings2 = getSignal2(
"00204143607", filePath="../signals/Hamid/00MaHaLI1002041436/")
# sig2, force2, fs2, firings2 = getSignal(2, 30, "./signals/")
signalWindow, labelWindow = self.prepareData(trainSigs[0],
trainFirings[0])
for index in range(1, len(trainSigs)):
sig, label = self.prepareData(trainSigs[index],
trainFirings[index])
signalWindow = np.append(signalWindow, sig, axis=0)
labelWindow = np.append(labelWindow, label, axis=0)
signalWindow2, labelWindow2 = self.prepareData(sig2, firings2)
X_train, X_test, Y_train, Y_test = trainTestSplit(
signalWindow, labelWindow, 0.75)
X_train2, X_test2, Y_train2, Y_test2 = trainTestSplit(
signalWindow2, labelWindow2, 0.75)
X_train = np.reshape(X_train, [-1, X_train.shape[1], X_train.shape[2]])
# X_test = np.reshape(X_test, [-1, X_test.shape[1], X_test.shape[2], 1])
Y_train = np.reshape(Y_train, [-1, Y_train.shape[1], Y_train.shape[2]])
# y_test = np.reshape(Y_test, [-1, Y_test.shape[1], Y_test.shape[2], 1])
# Y_train = (Y_train - np.min(Y_train)) / (np.max(Y_train) - np.min(Y_train))
# X_train = (X_train - np.min(X_train)) / (np.max(X_train) - np.min(X_train))
cycleGAN = CycleGANModified(Y_train.shape[1], Y_train.shape[2])
cycleGAN.train(x_train=X_train[:, :, range(Y_train.shape[2])],
y_train=Y_train,
epochs=100)
def test_train_test_split(self):
sig, force, fs, firings = getSignal(1, 30, "../signals/")
preparedFirings = prepareFiringSignal(firings[0], len(sig[0, 0][0]))
sdSig = calculateSD(sig)
signalWindow, labelWindow = windowingSig(sdSig, preparedFirings, windowSize=15)
X_train, X_test, y_train, y_test = trainTestSplit(signalWindow, labelWindow, 0.7)
TestCase.assertTrue(self, len(X_train) > 0)
TestCase.assertTrue(self, len(X_test) > 0)
TestCase.assertTrue(self, len(y_train, ) > 0)
TestCase.assertTrue(self, len(y_test) > 0)
signalWindow, labelWindow = prepareDataWithForce(trainSigs[0], trainTarget[0])
for index in range(1, len(trainSigs)):
sig, label = prepareDataWithForce(trainSigs[index], trainTarget[index])
signalWindow = np.append(signalWindow, sig, axis=0)
labelWindow = np.append(labelWindow, label, axis=0)
signalWindow2, labelWindow2 = prepareDataWithForce(sig2, force2)
# labelWindow[np.where(labelWindow >= 1)] = 1
# labelWindow2[np.where(labelWindow2 >= 1)] = 1
# num_classes = len(np.unique(labelWindow))
# labelWindow2[np.where(labelWindow2 > num_classes)] = num_classes - 1
labelWindow = labelWindow / np.max(labelWindow)
labelWindow2 = labelWindow2 / np.max(labelWindow2)
X_train, X_test, Y_train, Y_test = trainTestSplit(signalWindow, labelWindow,
0.75)
X_train2, X_test2, Y_train2, Y_test2 = trainTestSplit(signalWindow2,
labelWindow2, 0.75)
# Y_train = to_categorical(Y_train)
# Y_test = to_categorical(Y_test)
# Y_train2 = to_categorical(Y_train2)
# Y_test2 = to_categorical(Y_test2)
signal_size_row = X_train[0].shape[0]
signal_size_col = X_train[0].shape[1]
components = X_train[0].shape[2]
# X_train = np.reshape(X_train, [-1, signal_size_row, signal_size_col, 1])
# X_test = np.reshape(X_test, [-1, signal_size_row, signal_size_col, 1])
# X_train2 = np.reshape(X_train2, [-1, signal_size_row, signal_size_col, 1])
# X_test2 = np.reshape(X_test2, [-1, signal_size_row, signal_size_col, 1])
from keras.layers.advanced_activations import LeakyReLU
import numpy as np
from Utils.SignalUtils import getSignal, calculateSD, prepareFiringSignal, windowingSig, trainTestSplit, \
calculateCdr, calculateNormalizedOnCorr
sig, force, fs, firings = getSignal(1, 30, "../signals/")
sdSig = calculateSD(sig)
# sdSig = calculateICA(sdSig, 64)
sizeInputSignal = sdSig.shape[1]
preparedFirings = prepareFiringSignal(firings[0], sizeInputSignal)
signalWindow, labelWindow = windowingSig(sdSig, preparedFirings, windowSize=8) # should be divideable to 4
labelWindow = calculateCdr(labelWindow)
signalWindow = calculateNormalizedOnCorr(signalWindow)
x_train, x_test, y_train, y_test = trainTestSplit(signalWindow, labelWindow, 0.7)
# convert to one-hot vector
# y_train = to_categorical(y_train)
# y_test = to_categorical(y_test)
signal_size_row = x_train[0].shape[0]
signal_size_col = x_train[0].shape[1]
x_train = np.reshape(x_train, [-1, signal_size_row, signal_size_col])
x_test = np.reshape(x_test, [-1, signal_size_row, signal_size_col])
input_shape = (signal_size_row, signal_size_col)
model = Sequential()
model.add(Conv1D(16, 5, padding='same', input_shape=input_shape))
model.add(LeakyReLU())
model.add(Conv1D(8, 10, padding='same'))
model.add(LeakyReLU())
def test_plotForcePredicted(self):
maxFeatures = 64
# sig, force, fs, firings = getSignal(1, 30, "../signals/")
trainSigs = []
trainTargets = []
for i in range(1, 3):
sig, force, fs, firings = getSignal3(str(i))
trainSigs.append(sig)
trainTargets.append(force)
sig2, force2, fs2, firings2 = getSignal3(str(3))
# sig2, force2, fs2, firings2 = getSignal(2, 30, "./signals/")
def prepareDataWithForce(sig, force):
sdSig = calculateSD(sig)
sdSig = butter_bandpass_filter(sdSig, 20, 450, fs2)
# sdSig = calculateICA(sdSig, 64)
signalWindow, labelWindow = windowingSig(
sdSig, force,
windowSize=maxFeatures) # should be divideable to 4
# signalWindow = calculateFFTOnWindows(signalWindow)
labelWindow = calculateForceAverage(labelWindow)
# signalWindow, labelWindow = calculateNormalizedOnCorr(signalWindow, labelWindow)
# signalWindow, labelWindow = calculateICA(signalWindow, labelWindow, 4)
# signalWindow = extractPhaseSpace(signalWindow)
# signalWindow = extractSoundFeatures(signalWindow)
return signalWindow, labelWindow
def prepareData(sig, firings):
sdSig = calculateSD(sig)
# sdSig = calculateICA(sdSig, 64)
sdSig = butter_bandpass_filter(sdSig, 20, 450, 2048)
sizeInputSignal = sdSig.shape[1]
preparedFirings = prepareFiringSignal(firings[0], sizeInputSignal)
signalWindow, labelWindow = windowingSig(
sdSig, preparedFirings,
windowSize=maxFeatures) # should be divideable to 4
# signalWindow = calculateFFTOnWindows(signalWindow)
labelWindow = calculateCdr(labelWindow)
# signalWindow, labelWindow = calculateNormalizedOnCorr(signalWindow, labelWindow)
# signalWindow, labelWindow = calculateICA(signalWindow, labelWindow, 4)
signalWindow = extractPhaseSpace(signalWindow)
# signalWindow = extractSoundFeatures(signalWindow)
return signalWindow, labelWindow
signalWindow, labelWindow = prepareDataWithForce(
trainSigs[0], trainTargets[0])
for index in range(1, len(trainSigs)):
sig, label = prepareDataWithForce(trainSigs[index],
trainTargets[index])
signalWindow = np.append(signalWindow, sig, axis=0)
labelWindow = np.append(labelWindow, label, axis=0)
signalWindow2, labelWindow2 = prepareDataWithForce(sig2, force2)
# num_classes = len(np.unique(labelWindow))
# labelWindow2[np.where(labelWindow2 > num_classes)] = num_classes - 1
labelWindow = np.abs(labelWindow)
labelWindow2 = np.abs(labelWindow2)
labelWindow = (labelWindow - np.min(labelWindow)) / (
np.max(labelWindow) - np.min(labelWindow))
labelWindow2 = (labelWindow2 - np.min(labelWindow2)) / (
np.max(labelWindow2) - np.min(labelWindow2))
X_train, X_test, Y_train, Y_test = trainTestSplit(signalWindow,
labelWindow,
0.95,
shuffle=False)
X_train2, X_test2, Y_train2, Y_test2 = trainTestSplit(signalWindow2,
labelWindow2,
0.95,
shuffle=False)
# Y_train = to_categorical(Y_train)
# Y_test = to_categorical(Y_test)
# Y_train2 = to_categorical(Y_train2)
# Y_test2 = to_categorical(Y_test2)
signal_size_row = X_train[0].shape[0]
signal_size_col = X_train[0].shape[1]
x_train = np.reshape(X_train, [-1, signal_size_row, signal_size_col])
x_test = np.reshape(X_test, [-1, signal_size_row, signal_size_col])
X_train2 = np.reshape(X_train2, [-1, signal_size_row, signal_size_col])
X_test2 = np.reshape(X_test2, [-1, signal_size_row, signal_size_col])
input_shape = (signal_size_row, signal_size_col)
model = tf.keras.models.load_model(
'E:\Workspaces\crDsEMGEstimationPaper\EMGDecompositionDeepLearning\DeepLeraning\ForceEstimation.h5'
)
batch_size = 12
out = model.predict(X_train2, batch_size=batch_size)
rect_predict = out.ravel()
plt.plot(labelWindow2)
plt.plot(rect_predict)
plt.legend(['Target', 'Estimated'], loc='upper left')
plt.show()