# loadpath=loadpath, appoint=session)
testData = numpy.load(loadpath + '%s-Session%d-Data.npy' %
(gender, session))
testLabel = numpy.load(loadpath + '%s-Session%d-Label.npy' %
(gender, session))
testSeq = numpy.load(loadpath + '%s-Session%d-Seq.npy' %
(gender, session))
for episode in range(100):
graph = tensorflow.Graph()
with graph.as_default():
classifier = CTC_Multi_BLSTM(trainData=None,
trainLabel=None,
trainSeqLength=None,
featureShape=numpy.shape(
testData[0])[1],
numClass=5,
rnnLayers=2,
graphRevealFlag=False,
startFlag=False)
print('\nEpisode %d/100' % episode)
classifier.Load(loadpath=netpath %
(part, session, episode))
matrixDecode, matrixLogits, matrixSoftMax = classifier.Test_AllMethods(
testData=testData,
testLabel=testLabel,
testSeq=testSeq)
print('\n\n')
print(matrixDecode)
print(matrixLogits)
print(matrixSoftMax)
import os
if __name__ == '__main__':
part = 'Bands-30'
session = 6
startPosition = 13
loadpath = 'E:/CTC_Target_MSP/Feature/%s/' % part
savepath = 'E:/CTC_Target_MSP/CTC-MSP-Origin/%s-Session-%d/' % (part,
session)
trainData, trainLabel, trainSeq, trainScription, testData, testlabel, testSeq, testScription = Load_MSP(
loadpath=loadpath, appointSession=session)
graph = tensorflow.Graph()
with graph.as_default():
classifier = CTC_Multi_BLSTM(trainData=trainData,
trainLabel=trainScription,
trainSeqLength=trainSeq,
featureShape=len(trainData[0][0]),
numClass=5,
rnnLayers=2,
graphRevealFlag=False)
print(classifier.information)
classifier.Load(loadpath=savepath + '%04d-Network' % startPosition)
for episode in range(startPosition + 1, 100):
print('\nEpisode %d/100 : Total Loss = %f\n' %
(episode, classifier.Train()),
end='')
classifier.Save(savepath=savepath + '%04d-Network' % episode)
# exit()
if os.path.exists(savepath): exit()
os.makedirs(savepath + 'Decode')
os.makedirs(savepath + 'Logits')
os.makedirs(savepath + 'SoftMax')
trainData, trainLabel, trainSeq, trainScription, testData, testLabel, testSeq, testScription = Load_FAU(
loadpath=loadpath)
for episode in range(100):
graph = tensorflow.Graph()
with graph.as_default():
classifier = CTC_Multi_BLSTM(trainData=None,
trainLabel=None,
trainSeqLength=None,
featureShape=bands,
numClass=6,
rnnLayers=2,
graphRevealFlag=False,
startFlag=False)
print('\nEpisode %d/100' % episode)
classifier.Load(loadpath=netpath % (bands, episode))
matrixDecode, matrixLogits, matrixSoftMax = classifier.Test_AllMethodsWithLen(
testData=testData, testLabel=testLabel, testSeq=testSeq)
print('\n\n')
print(matrixDecode)
print(matrixLogits)
print(matrixSoftMax)
with open(savepath + 'Decode/%04d.csv' % episode, 'w') as file:
for indexX in range(5):
for indexY in range(5):
from CTC_Target.Loader.IEMOCAP_Loader import Load_MSP
import tensorflow
from CTC_Target.Model.CTC_Multi_BLSTM import CTC_Multi_BLSTM
import os
if __name__ == '__main__':
for part in ['Bands-30']:
loadpath = 'E:/CTC_Target_MSP/Feature/%s/' % part
for session in range(6, 7):
savepath = 'E:/CTC_Target_MSP/CTC-MSP-Origin/%s-Session-%d/' % (part, session)
if os.path.exists(savepath): continue
os.makedirs(savepath)
trainData, trainLabel, trainSeq, trainScription, testData, testlabel, testSeq, testScription = Load_MSP(
loadpath=loadpath, appointSession=session)
graph = tensorflow.Graph()
with graph.as_default():
classifier = CTC_Multi_BLSTM(trainData=trainData, trainLabel=trainScription, trainSeqLength=trainSeq,
featureShape=len(trainData[0][0]), numClass=5, rnnLayers=2,
graphRevealFlag=False)
print(classifier.information)
for episode in range(100):
print('\nEpisode %d/100 : Total Loss = %f\n' % (episode, classifier.Train()), end='')
classifier.Save(savepath=savepath + '%04d-Network' % episode)
# exit()