def main():
print("Initializing")
myAnn = keras_ann()
myloc = os.path.expanduser('~') + "/kerasTimeSeries/myweights/rnn"
myloc = '/nfshome/gst2d/localstorage/kerasTimeSeries/myweights/rnn'
myData = cnn_data(dataPath= os.path.expanduser('~') + "/eegData/")
tloc = myloc
myloc += '/'
for fname in listdir(myloc):
if ('fileModel' in fname and '~' not in fname):
useCandidate = myloc+str(fname)
testing = True
modelArgs = [] #getModels() small models only for now!
#addToModels(modelArgs)
print(useCandidate)
mod = pd.read_csv(useCandidate, sep='|', header=0)
print(mod.columns)
for index, candidate in pd.read_csv(useCandidate, sep='|', header=0).iterrows():
modelArgs.append(json.loads(candidate["model"]))
print(f"Number of Models: {len(modelArgs)}")
weights = []
myAnn.getWeights(weights,myloc)
print(weights)
myAnn.updatePaths(outputPath = os.path.dirname(os.path.realpath(__file__)) + "/")
myData.readData(fnames=inputData())
myAnn.testModel(modelArgs,myData.data,myData.labels,weights=weights,loadLoc=myloc)
def main():
# load models and weights
print("Initializing")
myAnn = keras_ann()
mySaveLoc = path.expanduser('~') + "/eegData/"
myLoadLoc = path.expanduser(
'~') + "/localstorage/kerasTimeSeries/myweights/"
myData = ann_data(dataPath=path.expanduser('~') + "/eegData/")
allWeights = []
models = []
freqs = ['delta', 'theta', 'alpha', 'beta1', 'beta2']
# there are several models to choose from,
# manually choose the index of the model
modelChoice = {}
for freq in freqs:
modelChoice[freq] = 0
modelChoice['delta'] = 1
for freq in freqs:
weights = []
myAnn.getWeights(weights, myLoadLoc + freq)
allWeights.append(myLoadLoc + freq + '/' +
weights[modelChoice[freq]][0])
#print(weights)
#print(weights[modelChoice[freq]][0])
for filename in listdir(myLoadLoc + freq):
if 'Model' in filename:
with open(myLoadLoc + freq + '/' + filename) as tfile:
modelIdtemp = 0
while modelIdtemp < modelChoice[freq]:
tfile.readline()
modelIdtemp += 1
#print(f"model: {modelIdtemp}")
models.append(
json.loads(tfile.readline().split('|')[-1].strip()))
#print(allWeights[-1], models[-1], '\n')
t = "input081.csv,input091.csv,input011.csv,input162.csv,input002.csv,input171.csv,input151.csv,input031.csv,input041.csv,input152.csv,input142.csv,input101.csv,input042.csv,input012.csv,input032.csv,input112.csv,input161.csv,input001.csv,input082.csv,input172.csv".split(
",")
i = 0
for inputfile in t:
print()
print('=' * 16)
print(inputfile)
myData.readData(fnames=[inputfile])
[normSTD, normMean] = myAnn.getNorm(myLoadLoc + freq + '/')
myData.expandDims()
myData.normalize(normSTD=normSTD, normMean=normMean)
myAnn.saveModelOutput(models,
myData.data,
myData.labels,
weights=allWeights,
saveLoc=mySaveLoc,
saveName='out' + inputfile,
loadLoc='')
i += 1
def main():
print("Initializing")
myAnn = keras_ann()
myloc = os.path.expanduser('~') + "/kerasTimeSeries/"
myData = cnn_data(dataPath= os.path.expanduser('~') + "/eegData/")
# 0 1 2 3
useCandidate = ['', 'topTwo.csv', 'topTen.csv', 'candidate.csv'][0]
testing = True
optimizeOptimizer = False
saveModel = False
modelArgs = [] #getModels() small models only for now!
#addToModels(modelArgs)
print("Collecting Models")
if (useCandidate == ''):
addToModels(modelArgs)
#else:
# getCandidates(modelArgs, fname=useCandidate, optimize = optimizeOptimizer) #
print(f"Number of Models: {len(modelArgs)}")
myAnn.updatePaths(outputPath = os.path.dirname(os.path.realpath(__file__)) + "/")
if (testing):
myData.readData()
else:
myData.readData(fnames=inputData())
lowFreq=highFreq=None
dataFiles = ",".join(inputData())
cvFolds = 0 if saveModel else 10
valPerc = 0.10
epochs = 1 if saveModel else 100
batchSize = 32 if saveModel else int(((myData.record_count*(1-valPerc))/cvFolds)+1)
with open("fileTrainTestParams.txt",'w') as params:
params.write(f"dataFiles: {dataFiles}\ncvFolds: {cvFolds}\n")
params.write(f"validation_split: {valPerc}\nepoch: {epochs}\n")
params.write(f"batchSize: {batchSize}\n")
params.write(f"frequency: {lowFreq} - {highFreq}\n")
params.write(f"normSTD : {myData.normSTD}\n")
params.write(f"normMean : {myData.normMean}")
if (saveModel):
myAnn.trainModel(modelArgs,myData.data,myData.labels, valSplit=valPerc, epochs=epochs, batchSize=batchSize, visualize=False, saveLoc=myloc)
return
if (testing):
myAnn.parameterSearch(modelArgs[:1],myData.data,myData.labels,valSplit=0.10)
else:
myAnn.parameterSearch(modelArgs,myData.data,myData.labels,numSplits=cvFolds, valSplit=valPerc, epochs=epochs, batchSize=batchSize, saveModel=saveModel, visualize=False, saveLoc=myloc)
def main():
if (len(sys.argv) < 2):
freqBands = ['theta']
else:
freqBands = []
for i in range(1, len(sys.argv)):
freqBands.append(sys.argv[i])
print("Initializing")
myAnn = keras_ann()
myloc = os.path.expanduser('~') + "/kerasTimeSeries/"
myData = ann_data(dataPath=os.path.expanduser('~') + "/eegData/")
testing = False
if (testing):
myData.readData()
else:
myData.readData(fnames=inputData())
for freqBand in freqBands:
print(f"FREQUENCY: {freqBand}")
weightPath = os.path.expanduser(
'~') + "/localstorage/kerasTimeSeries/myweights/" + freqBand + "/"
runTest(myAnn, myloc, weightPath, myData, freqBand)
def main():
if (len(sys.argv) < 2):
freqBand = 'theta'
else:
freqBand = sys.argv[1]
print("Initializing")
myAnn = keras_ann()
mybuild = ModelBuilder()
myloc = os.path.expanduser('~') + "/localstorage/kerasTimeSeries/"
weightPath=os.path.expanduser('~') + "/localstorage/kerasTimeSeries/myweights/" + freqBand + "/"
print("Collecting Models")
weights = []
modelArgs = []
print("GET PARAMS")
mybuild.getCandidates(modelArgs, fname=weightPath+"topTwo.csv", optimize = False)
myAnn.getWeights(weights,weightPath)
#=============
# for collecting data
#=============
print("GET PARAMS")
myData = ann_data(dataPath= os.path.expanduser('~') + "/eegData/")
[normSTD, normMean] = myAnn.getNorm(weightPath)
[lowFreq, highFreq, _] = ann_data.getFreqBand(freqBand)
testing = True
if (testing):
myData.readData()
else:
pass#myData.readData(fnames=inputData())
#myData.filterFrequencyRange(low=lowFreq, high=highFreq)
myData.expandDims()
myData.normalize(normSTD=normSTD, normMean=normMean)
#print(weights)
print("PRINT MODEL")
myAnn.printModel(modelArgs, weights=weights,printLoc=myloc, loadLoc=weightPath,X=myData.data,Y=myData.labels)
print("DONE")
def main():
if (len(sys.argv) < 2):
freqBand = 'theta'
else:
freqBand = sys.argv[1]
print("Initializing")
myAnn = keras_ann()
myloc = os.path.expanduser('~') + "/kerasTimeSeries/"
weightPath = os.path.expanduser(
'~') + "/localstorage/kerasTimeSeries/myweights/" + freqBand + "/"
myData = ann_data(dataPath=os.path.expanduser('~') + "/eegData/")
modelArgs = []
print("Collecting Models")
getCandidates(modelArgs, fname=weightPath + "topTwo.csv", optimize=False)
weights = []
myAnn.getWeights(weights, weightPath)
[normSTD, normMean] = myAnn.getNorm(weightPath)
[lowFreq, highFreq, _] = ann_data.getFreqBand(freqBand)
#print(modelArgs)
#print(weights)
#return
testing = False
if (testing):
myData.readData()
else:
myData.readData(fnames=inputData())
myData.filterFrequencyRange(low=lowFreq, high=highFreq)
myData.expandDims()
myData.normalize(normSTD=normSTD, normMean=normMean)
myAnn.testModel(modelArgs,
myData.data,
myData.labels,
weights=weights,
loadLoc=weightPath)
def main():
print("Initializing")
myAnn = keras_ann()
myloc = os.path.expanduser('~') + "/kerasTimeSeries/"
myData = ann_data(dataPath=os.path.expanduser('~') + "/eegData/")
#0 1 2 3 4
freqBand = ['delta', 'theta', 'alpha', 'beta1', 'beta2'][4]
[lowFreq, highFreq, kernelsize] = ann_data.getFreqBand(freqBand)
lowFreq = highFreq = None
# 0 1 2 3
useCandidate = ['', 'topTwo.csv', 'topTen.csv', 'candidate.csv'][3]
testing = True
optimizeOptimizer = False
saveModel = True
modelArgs = [] #getModels() small models only for now!
#addToModels(modelArgs)
print("Collecting Models")
if (useCandidate == ''):
addToModelsTest_FrequencyFilters(modelArgs,
addConvFilters=False,
manyFilters=False,
numKeepIndexes=100,
kernalPreset=kernelsize)
else:
getCandidates(modelArgs,
fname=useCandidate,
optimize=optimizeOptimizer) #
myAnn.updatePaths(outputPath=os.path.dirname(os.path.realpath(__file__)) +
"/")
if (testing):
myData.readData()
else:
myData.readData(fnames=inputData())
myData.filterFrequencyRange(low=lowFreq, high=highFreq)
myData.expandDims()
myData.normalize()
dataFiles = ",".join(inputData())
cvFolds = 0 if saveModel else 10
valPerc = 0.10
epochs = 1 if saveModel else 100
batchSize = 32 if saveModel else int(((myData.record_count *
(1 - valPerc)) / cvFolds) + 1)
with open("fileTrainTestParams.txt", 'w') as params:
params.write(f"dataFiles: {dataFiles}\ncvFolds: {cvFolds}\n")
params.write(f"validation_split: {valPerc}\nepoch: {epochs}\n")
params.write(f"batchSize: {batchSize}\n")
params.write(f"frequency: {lowFreq} - {highFreq}\n")
params.write(f"normSTD : {myData.normSTD}\n")
params.write(f"normMean : {myData.normMean}")
if (saveModel):
myAnn.trainModel(modelArgs,
myData.data,
myData.labels,
valSplit=valPerc,
epochs=epochs,
batchSize=batchSize,
visualize=False,
saveLoc=myloc)
return
if (testing):
myAnn.parameterSearch(modelArgs[:10],
myData.data,
myData.labels,
valSplit=0.10)
else:
myAnn.parameterSearch(modelArgs,
myData.data,
myData.labels,
numSplits=cvFolds,
valSplit=valPerc,
epochs=epochs,
batchSize=batchSize,
saveModel=saveModel,
visualize=False,
saveLoc=myloc)