def train(self, reader, doBalanceLabels):
trainFiles = reader.traindata
triggers = reader.classtrainTriggers
onsets = reader.classtrainOnsets
finOnsets = reader.classtrainFinOnsets
[self.trainData,
self.labels] = utilities.dataLoadFromEDF(self, trainFiles, triggers,
onsets, finOnsets,
self.params)
if (doBalanceLabels):
[self.trainData,
self.labels] = utilities.balance_labels(self.trainData,
self.labels)
numChannels = self.trainData.shape[1]
numSamples = self.trainData.shape[0]
fvecLen = len(self.params.channelSelect)
fVecs = np.zeros((numSamples, fvecLen))
for k in range(numSamples):
#print("Fvec: " + str(k) + " of " + str(numSamples))
for i in range(fvecLen):
sample = np.squeeze(self.trainData[k, i, :])
if not (self.params.fDiaps is None):
self.params.lowFreq = self.params.fDiaps[i][0]
self.params.highFreq = self.params.fDiaps[i][1]
specVal = np.mean(utilities.get1DFeatures(
sample, self.params)) #!!!!!!!
fVecs[k, i] = specVal
#Shuffle!
inds = np.random.permutation(fVecs.shape[0])
fVecs = fVecs[inds, :]
labels = self.labels[inds]
self.trainResult.mean = np.mean(fVecs, 0)
self.trainResult.std = np.std(fVecs, 0)
#Norm!
for i in range(fVecs.shape[0]):
fVecs[i, :] = (fVecs[i, :] -
self.trainResult.mean) / self.trainResult.std
fTransformed = fVecs
# PCA!
if self.params.usePCA:
pcaTransform = PCA(self.params.numPC)
pcaTransform.fit(fVecs)
self.trainResult.pcaOp = pcaTransform
fTransformed = pcaTransform.transform(fVecs)
#LDA!
if not (self.params.finalClassifier is None):
[Op, fTransformed
] = utilities.trainClassifier(self.params.finalClassifier,
fTransformed, labels)
self.trainResult.finalOp = Op
if not (self.params.distanceFun is None):
self.trainResult.trainTransformedVecs = fTransformed
self.trainResult.trainLabels = labels
def train(self,trainFiles,triggers,onsets,finOnsets, doBalanceLabels):
[self.trainData, self.labels] = utilities.dataLoadFromEDF(self, trainFiles, triggers, onsets, finOnsets,self.params)
self.numClasses = len(set(self.labels))
if (doBalanceLabels):
[self.trainData, self.labels] = utilities.balance_labels(self.trainData, self.labels)
cspModels = []
for i in range(self.numClasses):
for j in range(self.numClasses):
if i<j:
print("Training SWCSP Model for {} vs {}".format(i,j))
classFts1 = self.trainData[self.labels == i, :, :]
classFts2 = self.trainData[self.labels == j, :, :]
S = []
S.append(classFts1)
S.append(classFts2)
cspWorker = SWCSP.SWCSP(self.params.Fs, self.numCSP)
cspWorker.train(S)
cspModels.append(cspWorker)
fVecs = np.zeros((self.labels.shape[0],self.numCSP*2*len(cspModels)))
for i in range(self.labels.shape[0]):
fvec=[]
for j in range(len(cspModels)):
fvec = np.concatenate((fvec, cspModels[j].process(self.trainData[i,:])))
fVecs[i,:] = fvec
self.trainResult.cspOp = cspModels
#Shuffle!
inds = np.random.permutation(fVecs.shape[0])
fVecs = fVecs[inds,:]
labels = self.labels[inds]
self.trainResult.mean = np.mean(fVecs,0)
self.trainResult.std = np.std(fVecs,0)
#Norm!
for i in range(fVecs.shape[0]):
fVecs[i,:] = (fVecs[i,:]-self.trainResult.mean)/self.trainResult.std
fTransformed = fVecs
# LDA!
if not (self.params.finalClassifier is None):
[Op, fTransformed] = utilities.trainClassifier(self.params.finalClassifier, fTransformed, labels)
self.trainResult.finalOp = Op
if not (self.params.distanceFun is None):
self.trainResult.trainTransformedVecs = fTransformed
self.trainResult.trainLabels = labels
def train(self, reader, doBalanceLabels):
trainFiles = reader.traindata
triggers = reader.classtrainTriggers
onsets = reader.classtrainOnsets
finOnsets = reader.classtrainFinOnsets
[self.trainData,
self.labels] = utilities.dataLoadFromEDF(self, trainFiles, triggers,
onsets, finOnsets,
self.params)
self.numClasses = len(set(self.labels))
if (doBalanceLabels):
[self.trainData,
self.labels] = utilities.balance_labels(self.trainData,
self.labels)
numSamples = self.trainData.shape[0]
tmpVec = utilities.get2DFeatures(np.squeeze(self.trainData[0, :, :]),
self.params)
if (np.ndim(tmpVec) == 3):
fVecs = np.zeros((numSamples, tmpVec.shape[0], tmpVec.shape[1],
tmpVec.shape[2]))
if (np.ndim(tmpVec) == 2):
fVecs = np.zeros((numSamples, tmpVec.shape[0], tmpVec.shape[1]))
for k in range(numSamples):
#print('Train Sample ', k, ' of ', numSamples)
curFvec = utilities.get2DFeatures(
np.squeeze(self.trainData[k, :, :]), self.params)
if (np.ndim(tmpVec) == 3):
fVecs[k, :, :, :] = curFvec
if (np.ndim(tmpVec) == 2):
fVecs[k, :, :, ] = curFvec
#Shuffle!
inds = np.random.permutation(fVecs.shape[0])
if (np.ndim(tmpVec) == 3):
fVecs = fVecs[inds, :, :, :]
if (np.ndim(tmpVec) == 2):
fVecs = fVecs[inds, :, :]
labels = self.labels[inds]
# ConvNet!
model = self.neuralFun(self.numClasses)
model.trainModel(fVecs, labels, self.batchSize, self.numEpochs)
self.trainResult.model = model
def train(self, reader, doBalanceLabels):
trainFiles = reader.traindata
triggers = reader.classtrainTriggers
onsets = reader.classtrainOnsets
finOnsets = reader.classtrainFinOnsets
[self.trainData,
self.labels] = utilities.dataLoadFromEDF(self, trainFiles, triggers,
onsets, finOnsets,
self.params)
if (doBalanceLabels):
[self.trainData,
self.labels] = utilities.balance_labels(self.trainData,
self.labels)
csp = CSP(n_components=self.numCSP,
reg=None,
log=True,
norm_trace=False)
csp.fit(self.trainData, self.labels)
fVecs = csp.transform(self.trainData)
self.trainResult.cspOp = csp
#Shuffle!
inds = np.random.permutation(fVecs.shape[0])
fVecs = fVecs[inds, :]
labels = self.labels[inds]
self.trainResult.mean = np.mean(fVecs, 0)
self.trainResult.std = np.std(fVecs, 0)
#Norm!
for i in range(fVecs.shape[0]):
fVecs[i, :] = (fVecs[i, :] -
self.trainResult.mean) / self.trainResult.std
fTransformed = fVecs
#LDA!
if not (self.params.finalClassifier is None):
[Op, fTransformed
] = utilities.trainClassifier(self.params.finalClassifier,
fTransformed, labels)
self.trainResult.finalOp = Op
self.trainResult.trainTransformedVecs = fTransformed
self.trainResult.trainLabels = labels
def train(self, reader, doBalanceLabels):
trainFiles = reader.traindata
triggers = reader.classtrainTriggers
onsets = reader.classtrainOnsets
finOnsets = reader.classtrainFinOnsets
[self.trainData,
self.labels] = utilities.dataLoadFromEDF(self, trainFiles, triggers,
onsets, finOnsets,
self.params)
if (doBalanceLabels):
[self.trainData,
self.labels] = utilities.balance_labels(self.trainData,
self.labels)
numChannels = self.trainData.shape[1]
numSamples = self.trainData.shape[0]
sumFvecLen = 0
for ch in range(numChannels):
if not (self.params.fDiaps is None):
self.params.lowFreq = self.params.fDiaps[ch][0]
self.params.highFreq = self.params.fDiaps[ch][1]
tmpVec = utilities.get1DFeatures(
np.squeeze(self.trainData[0, ch, :]), self.params)
sumFvecLen = sumFvecLen + len(tmpVec)
fVecs = np.zeros((numSamples, sumFvecLen))
for k in range(numSamples):
#print('Train Sample ', k, ' of ', numSamples)
curFvec = np.zeros((0))
for i in range(numChannels):
if not (self.params.fDiaps is None):
self.params.lowFreq = self.params.fDiaps[i][0]
self.params.highFreq = self.params.fDiaps[i][1]
chVec = utilities.get1DFeatures(
np.squeeze(self.trainData[k, i, :]), self.params)
curFvec = np.concatenate((curFvec, chVec), axis=0)
fVecs[k, :] = curFvec
#Shuffle!
inds = np.random.permutation(fVecs.shape[0])
fVecs = fVecs[inds, :]
labels = self.labels[inds]
self.trainResult.mean = np.mean(fVecs, 0)
self.trainResult.std = np.std(fVecs, 0)
#Norm!
for i in range(fVecs.shape[0]):
fVecs[i, :] = (fVecs[i, :] -
self.trainResult.mean) / self.trainResult.std
# PCA!
fTransformed = fVecs
if self.params.usePCA:
pcaTransform = PCA(self.params.numPC)
pcaTransform.fit(fVecs)
self.trainResult.pcaOp = pcaTransform
fTransformed = pcaTransform.transform(fVecs)
#LDA!
if not (self.params.finalClassifier is None):
[Op, fTransformed
] = utilities.trainClassifier(self.params.finalClassifier,
fTransformed, labels)
self.trainResult.finalOp = Op
self.trainResult.trainLabels = labels
if not (self.params.distanceFun is None):
self.trainResult.trainTransformedVecs = fTransformed
def train(self, reader, doBalanceLabels):
trainFiles = reader.traindata
triggers = reader.classtrainTriggers
onsets = reader.classtrainOnsets
finOnsets = reader.classtrainFinOnsets
[self.trainData,
self.labels] = utilities.dataLoadFromEDF(self, trainFiles, triggers,
onsets, finOnsets,
self.params)
if (doBalanceLabels):
[self.trainData,
self.labels] = utilities.balance_labels(self.trainData,
self.labels)
numChannels = self.trainData.shape[1]
numSamples = self.trainData.shape[0]
tmpVec = utilities.get1DFeatures(np.squeeze(self.trainData[0, 0, :]),
self.params)
fvecLen = len(tmpVec)
fVecs = np.zeros((numSamples, numChannels, fvecLen))
for k in range(numSamples):
#print("Fvec: " + str(k) + " of " + str(numSamples))
for i in range(numChannels):
fVecs[k, i, :] = utilities.get1DFeatures(
np.squeeze(self.trainData[k, i, :]), self.params)
#Shuffle!
inds = np.random.permutation(fVecs.shape[0])
fVecs = fVecs[inds, :, :]
labels = self.labels[inds]
#Norm!
for i in range(fVecs.shape[1]):
chanVecs = np.squeeze(fVecs[:, i, :])
self.trainResult.mean.append(np.mean(chanVecs, 0))
self.trainResult.std.append(np.std(chanVecs, 0))
for k in range(chanVecs.shape[0]):
chanVecs[k, :] = (chanVecs[k, :] - self.trainResult.mean[i]
) / self.trainResult.std[i]
fVecs[:, i, :] = chanVecs
fTransformed = np.zeros(
(fVecs.shape[0], fVecs.shape[1] * fVecs.shape[2]))
if (self.params.usePCA):
fTransformed = np.zeros(
(fVecs.shape[0], fVecs.shape[1] * self.params.numPC))
# PCA or reshaping:
for i in range(numChannels):
chanVecs = np.squeeze(fVecs[:, i, :])
if (self.params.usePCA):
curPcaTransform = PCA(self.params.numPC)
curPcaTransform.fit(chanVecs)
self.trainResult.pcaOp.append(curPcaTransform)
fTransformed[:, i * self.params.numPC:(i + 1) *
self.params.numPC] = curPcaTransform.transform(
chanVecs)
else:
fTransformed[:, i * fVecs.shape[2]:(i + 1) *
fVecs.shape[2]] = chanVecs
#LDA!
if not (self.params.finalClassifier is None):
[Op, fTransformed
] = utilities.trainClassifier(self.params.finalClassifier,
fTransformed, labels)
self.trainResult.finalOp = Op
if not (self.params.distanceFun is None):
self.trainResult.trainTransformedVecs = fTransformed
self.trainResult.trainLabels = labels
def train(self, reader, doBalanceLabels):
trainFiles = reader.traindata
triggers = reader.classtrainTriggers
onsets = reader.classtrainOnsets
finOnsets = reader.classtrainFinOnsets
[self.trainData, self.labels] = utilities.dataLoadFromEDF(self, trainFiles, triggers, onsets, finOnsets,
self.params)
if (doBalanceLabels):
[self.trainData, self.labels] = utilities.balance_labels(self.trainData, self.labels)
numChannels = self.trainData.shape[1]
numSamples = self.trainData.shape[0]
if self.mode == 'IndepChan':
tmpVec = utilities.get1DFeatures(np.squeeze(self.trainData[0, 0, :]), self.params)
sumFvecLen = len(tmpVec)
print(sumFvecLen, self.trainData.shape)
print(self.trainData[0, 0, :])
if self.mode == 'CustomFreqs':
sumFvecLen = len(self.params.channelSelect)
print(sumFvecLen, '\n', self.params.channelSelect)
if self.mode == 'Classic':
sumFvecLen = 0
for i in range(numChannels):
if not (self.params.fDiaps is None):
self.params.lowFreq = self.params.fDiaps[i][0]
self.params.highFreq = self.params.fDiaps[i][1]
tmpVec = utilities.get1DFeatures(np.squeeze(self.trainData[0, i, :]), self.params)
sumFvecLen = sumFvecLen + len(tmpVec)
#print(sumFvecLen, '\n', self.trainData[0, i, :], '\n', tmpVec)
fVecs = np.zeros((numSamples, numChannels, sumFvecLen))
#Fastovets
#indepchan [nchan = 75 x nvecs = 335 x vecLen = 49] 139
#classic [nchan = 75 x nvecs = 335 x vecLen = 49] 2293
#custom [nchan = 75 x nvecs = 335 x vecLen = 29]
#tmpVec.shape = 16
for k in range(numSamples): #cust classic = 1
curFvec = np.zeros((0))
# print("Fvec: " + str(k) + " of " + str(numSamples))
for i in range(numChannels):
if not (self.params.fDiaps is None):
self.params.lowFreq = self.params.fDiaps[i][0]
self.params.highFreq = self.params.fDiaps[i][1]
tmpVec = utilities.get1DFeatures(np.squeeze(self.trainData[k, i, :]), self.params)
if self.mode == 'IndepChan':
fVecs[k, i, :] = tmpVec #"""КАК НАСЧЕТ ТОГО, ЧТОБЫ СОБЛЮДАТЬ РАЗМЕРНОСТЬ FVECS?""" [k, i, :]
if self.mode == 'CustomFreqs':
specVal = np.mean(tmpVec) # !!!!!!! was not meaned while sumfveclen was counted!
fVecs[k, i, 0] = specVal #[k, i]
if self.mode == 'Classic':
curFvec = np.concatenate((curFvec, tmpVec), axis=0)
if self.mode == 'Classic':
fVecs[k, i, :] = curFvec #[k, :]
# Shuffle!
inds = np.random.permutation(fVecs.shape[0])
fVecs = fVecs[inds, :, :]
labels = self.labels[inds]
#print(self.trainResult.mean, self.trainResult.std)
# Norm!
if self.mode == 'Classic' or self.mode == 'CustomFreqs':
self.trainResult.mean = np.mean(fVecs, 0)
self.trainResult.std = np.std(fVecs, 0)
for i in range(fVecs.shape[0]):
fVecs[i, :, :] = (fVecs[i, :, :] - self.trainResult.mean) / self.trainResult.std
fTransformed = fVecs
#PCA
if self.params.usePCA: #cycle! as in INDEPCHAN
for i in range(fVecs.shape[1]):
tmp = np.zeros((fVecs.shape[0], fVecs.shape(2)))
tmp = fVecs(:, i, :)
pcaTransform = PCA(self.params.numPC)
pcaTransform.fit(tmp)
self.trainResult.pcaOp = pcaTransform
fTransformed = pcaTransform.transform(tmp)