def test_classifier(paramsClassifier, settingsClassifier, baselineValues):
'''Performs network testing.
Input:
-classifier parameters
-classifier settings values
-sensor baseline values
'''
nrInputNeurons = paramsClassifier['NUM_INPUT_NEURONS']
nrVR = paramsClassifier['NUM_VR']
alpha = paramsClassifier['RATE_ALPHA']
logFolder = settingsClassifier['TEST_LOGGER_FOLDER']
masterPath = settingsClassifier['MASTER_PATH']
spikeSourceVRTest = settingsClassifier['SPIKE_SOURCE_VR_RESPONSE_TEST']
classLabelsTestFile = settingsClassifier['CLASS_LABELS_TEST']
spikeLengthSample = settingsClassifier['SPIKE_TRAIN_LENGTH']
nrSamples = settingsClassifier['NUM_LOG_SAMPLES']
logPath = os.path.join(masterPath, logFolder)
files = os.listdir(logPath)
pattern = '*.csv'
fileList = [name for name in files if fnmatch.fnmatch(name, pattern)]
fileArray = np.ravel(np.array(fileList))
np.random.shuffle(fileArray)
odourClassesTest = [int(fileName[0]) for fileName in fileArray]
concatenate_recordings(logPath, fileArray, spikeSourceVRTest,
odourClassesTest, nrSamples, nrInputNeurons,
nrVR, spikeLengthSample, alpha, baselineValues)
make_labels_file(classLabelsTestFile, odourClassesTest)
settingsClassifier['LEARNING'] = False
eNoseClassifier.runClassifier(paramsClassifier, settingsClassifier, None)
def train_classifier(paramsClassifier, settingsClassifier, baselineValues):
'''Performs network training.
Input:
-classifier parameters
-classifier settings values
-sensor baseline values
Output:
-saves the trained PN-AN connection weights to file
'''
nrInputNeurons = paramsClassifier['NUM_INPUT_NEURONS']
nrVR = paramsClassifier['NUM_VR']
spikeLengthSample = settingsClassifier['SPIKE_TRAIN_LENGTH']
alpha = paramsClassifier['RATE_ALPHA']
logFolder = settingsClassifier['TRAIN_LOGGER_FOLDER']
masterPath = settingsClassifier['MASTER_PATH']
spikeSourceVRTrain = settingsClassifier['SPIKE_SOURCE_VR_RESPONSE_TRAIN']
classLabelsTrainFile = settingsClassifier['CLASS_LABELS_TRAIN']
classActivationsFile = settingsClassifier['SPIKE_SOURCE_CLASS_ACTIVATIONS']
nrRepetitions = settingsClassifier['NUM_REPETITIONS']
nrSamples = settingsClassifier['NUM_LOG_SAMPLES']
maxSimTime = nrSamples*spikeLengthSample
logPath = os.path.join(masterPath, logFolder)
files = os.listdir(logPath)
pattern = '*.csv'
fileList = [name for name in files if fnmatch.fnmatch(name, pattern)]
fileArray = np.ravel(fileList * nrRepetitions)
np.random.shuffle(fileArray)
odourClassesTrain = [int(fileName[0]) for fileName in fileArray]
concatenate_recordings(logPath, fileArray, spikeSourceVRTrain,
odourClassesTrain, nrSamples, nrInputNeurons,
nrVR, spikeLengthSample, alpha, baselineValues)
make_class_activation_spikes(classActivationsFile,
odourClassesTrain,
maxSimTime)
make_labels_file(classLabelsTrainFile, odourClassesTrain)
settingsClassifier['LEARNING'] = True
eNoseClassifier.runClassifier(paramsClassifier, settingsClassifier, None)
def cross_validate(paramsClassifier, settingsClassifier, baselineValues):
'''Performs crossvalidation on the data set.
Input:
-classifier parameters
-classifier settings values
-sensor baseline values
Output:
-crossvalidation network performance scores vector
'''
nrInputNeurons = paramsClassifier['NUM_INPUT_NEURONS']
nrVR = paramsClassifier['NUM_VR']
nrClasses = paramsClassifier['NUM_CLASSES']
alpha = paramsClassifier['RATE_ALPHA']
logFolder = settingsClassifier['CROSSVALIDATION_LOGGER_FOLDER']
nrFolds = settingsClassifier['NUM_FOLDS']
masterPath = settingsClassifier['MASTER_PATH']
spikeSourceVRTrain = settingsClassifier['SPIKE_SOURCE_VR_RESPONSE_TRAIN']
spikeSourceVRTest = settingsClassifier['SPIKE_SOURCE_VR_RESPONSE_TEST']
classLabelsTrainFile = settingsClassifier['CLASS_LABELS_TRAIN']
classLabelsTestFile = settingsClassifier['CLASS_LABELS_TEST']
classActivationsFile = settingsClassifier['SPIKE_SOURCE_CLASS_ACTIVATIONS']
nrSamples = settingsClassifier['NUM_LOG_SAMPLES']
spikeLengthSample = settingsClassifier['SPIKE_TRAIN_LENGTH']
maxSimTime = nrSamples*spikeLengthSample
logPath = os.path.join(masterPath, logFolder)
files = os.listdir(logPath)
fileList = [[]] * nrClasses
for cls in range(nrClasses):
pattern = str(cls)+'-*.csv'
tempFileList = [name for name in files if fnmatch.fnmatch(name, pattern)]
fileList[cls] = tempFileList
fileArray = np.array(fileList)
nrTotObs = np.shape(fileArray)[1]
scores = np.zeros((1, nrFolds))
for fold in range(nrFolds):
indices = range(0, nrTotObs)
indxTest = range(nrTotObs/nrFolds*fold, nrTotObs/nrFolds*(fold+1))
indxTrain = np.setdiff1d(indices,indxTest)
trainFiles = np.ravel(fileArray[:, indxTrain])
testFiles = np.ravel(fileArray[:, indxTest])
np.random.shuffle(trainFiles)
np.random.shuffle(testFiles)
odourClassesTrain = [int(fileName[0]) for fileName in trainFiles]
odourClassesTest = [int(fileName[0]) for fileName in testFiles]
concatenate_recordings(logPath, trainFiles, spikeSourceVRTrain,
odourClassesTrain, nrSamples, nrInputNeurons,
nrVR, spikeLengthSample, alpha, baselineValues)
concatenate_recordings(logPath, testFiles, spikeSourceVRTest,
odourClassesTest, nrSamples, nrInputNeurons,
nrVR, spikeLengthSample, alpha, baselineValues)
make_class_activation_spikes(classActivationsFile,
odourClassesTrain,
maxSimTime)
make_labels_file(classLabelsTrainFile, odourClassesTrain)
make_labels_file(classLabelsTestFile, odourClassesTest)
settingsClassifier['NUM_OBSERVATIONS'] = len(odourClassesTrain)
settingsClassifier['NUM_OBSERVATIONS_TEST'] = len(odourClassesTest)
settingsClassifier['LEARNING'] = True
eNoseClassifier.runClassifier(paramsClassifier, settingsClassifier, fold)
print 'Training completed'
raw_input("Press Enter to proceed to testing...")
settingsClassifier['LEARNING'] = False
scores[0, fold] = eNoseClassifier.runClassifier(paramsClassifier,
settingsClassifier, fold)
print 'Testing for fold ' + str(fold) + ' completed'
raw_input("Press Enter to continue with the next fold...")
return scores