train_fileTripletL = readTrainFileIDsUsedForTraining(params, classifierID)
train_fileIDs = [train_fileID for _, _, train_fileID in train_fileTripletL]
# print('# train_fileIDs =', train_fileIDs)
test_fileTripletL = getFilesNotUsedInTrain(params, train_fileIDs)
all_fileTripletL = test_fileTripletL
allFileIDandClassifierIDs = [(fileID, classifierID) for fileID in fileIDs]
print('# allFileIDandClassifierIDs =', allFileIDandClassifierIDs)
y_testL, y_predL = [], []
for fileCnt, fileIDandClassifierID in enumerate(allFileIDandClassifierIDs):
epochNumByStage_testL, epochNumByStage_predL, avg_ep_testL, avg_ep_predL = [], [], [], []
print('fileIDandClassifierID = ' + str(fileIDandClassifierID))
fileID = fileIDandClassifierID[0]
predictionTargetDataFilePath = params.pickledDir + '/' + params.eegFilePrefix + '.' + fileID + '.pkl'
print('predictionTargetDataFilePath =', predictionTargetDataFilePath)
dataFileHandler = open(predictionTargetDataFilePath, 'rb')
(eeg, ch2, stageSeq, timeStamps) = pickle.load(dataFileHandler)
totalEpochNum = len(stageSeq[lstm_length - 1:])
params.markovOrderForPrediction = 0
(y_test, y_pred) = classifySequentially(params, paramID,
params.pickledDir,
fileIDandClassifierID)
y_testL.append(y_test)
y_predL.append(y_pred)
allFileIDandClassifierIDsL.append(allFileIDandClassifierIDs)
y_predLL.append(y_predL)
with open('../data/pickled/y_test_and_y_pred_for_graphs.pkl', 'wb') as f:
pickle.dump((allFileIDandClassifierIDsL, y_testL, y_predLL), f)
testFileIDandClassifierIDs = [(test_fileID, classifierID) for test_fileID in test_fileIDs]
# print('# testFileIDandClassifierIDs =', testFileIDandClassifierIDs)
y_test_byFile, y_pred_byFile = [], []
for testFileCnt, testFileIDandClassifierID in enumerate(testFileIDandClassifierIDs):
epochNumByStage_testL, epochNumByStage_predL, avg_ep_testL, avg_ep_predL = [], [], [], []
print('testFileIDandClassifierID = ' + str(testFileIDandClassifierID))
testFileID = testFileIDandClassifierID[0]
# print('testFileIDandClassifierID[0] =', testFileIDandClassifierID[0])
predictionTargetDataFilePath = params.pickledDir + '/' + params.eegFilePrefix + '.' + testFileID + '.pkl'
print('predictionTargetDataFilePath =', predictionTargetDataFilePath)
dataFileHandler = open(predictionTargetDataFilePath, 'rb')
(eeg, ch2, stageSeq, timeStamps) = pickle.load(dataFileHandler)
totalEpochNum = len(stageSeq[lstm_length-1:])
params_for_classifier = ParameterSetup(paramFileName='params.'+classifierID+'.json')
params_for_classifier.markovOrderForPrediction = markovOrder
(y_test, y_pred) = classifySequentially(params_for_classifier, paramID, params.pickledDir, testFileIDandClassifierID)
y_test_byFile.append(y_test)
y_pred_byFile.append(y_pred)
# loopCnt += 1
# if loopCnt > 2:
# break
testFileIDandClassifierIDs_byBlock.append(testFileIDandClassifierIDs)
y_test_byBlock.append(y_test_byFile)
y_pred_byBlock.append(y_pred_byFile)
# if loopCnt > 2:
# break
testFileIDandClassifierIDs_byMethod.append(testFileIDandClassifierIDs_byBlock)
y_test_byMethod.append(y_test_byBlock)
y_pred_byMethod.append(y_pred_byBlock)
specificityMat = np.zeros((orderNum, labelNum), dtype=float)
accuracyMat = np.zeros((orderNum, labelNum), dtype=float)
precisions = np.zeros((orderNum), dtype=float)
for markovOrder in range(orderMin, orderMax + 1):
print('markovOrder = ' + str(markovOrder))
params.pastStageLookUpNum = markovOrder
for fileID in fileIDs:
print(' extracting features for fileID = ' + str(fileID))
featureExtraction(params, fileID)
for fileID in fileIDs:
print(' training and testing for fileID = ' + str(fileID))
trainClassifier(params, paramID, fileID)
(y_test, y_pred) = classifySequentially(params, paramID, paramDir,
fileID)
(stageLabels, sensitivity, specificity,
accuracy) = y2sensitivity(y_test, y_pred)
# (stageLabels4confusionMat, confusionMat) = y2confusionMat(y_test, y_pred)
# printConfusionMat(stageLabels4confusionMat, confusionMat)
y_matching = (y_test == y_pred)
correctNum = sum(y_matching)
# print('y_test = ' + str(y_test[:50]))
# print('y_pred = ' + str(y_pred[:50]))
y_length = y_pred.shape[0]
precision = correctNum / y_length
for labelID in range(labelNum):
print(' stageLabel = ' + stageLabels[labelID] +
', sensitivity = ' + "{0:.3f}".format(sensitivity[labelID]) +
', specificity = ' + "{0:.3f}".format(specificity[labelID]) +
def test_by_classifierID(params, datasetType, classifierID):
paramDir = params.pickledDir
testFileDir = params.pickledDir
stageLabels = params.stageLabels4evaluation
labelNum = len(stageLabels)
resultFileDescription = ''
paramID = 0
markovOrder = 0
fileTripletL = readTrainFileIDsUsedForTraining(params, classifierID)
train_fileIDs = [fileID for _, _, fileID in fileTripletL]
# print('# train_fileIDs =', train_fileIDs)
params_test = params
if datasetType == 'test':
params_test.pickledDir = testFileDir
test_fileTripletL = getFilesNotUsedInTrain(params_test, train_fileIDs)
testFileIDandClassifierIDs = [(test_fileID, classifierID) for _, _, test_fileID in test_fileTripletL]
fileNum = len(testFileIDandClassifierIDs)
print('# testFileIDandClassifierIDs =', testFileIDandClassifierIDs)
# totalConfusionMat = np.zeros((labelNum, labelNum))
# for paramID in range(len(classifierParams)):
# print('classifier parameter = ' + str(classifierParams[paramID]))
sensitivityL = [[] for _ in range(labelNum)]
specificityL = [[] for _ in range(labelNum)]
accuracyL = [[] for _ in range(labelNum)]
precisionL = [[] for _ in range(labelNum)]
f1scoreL = [[] for _ in range(labelNum)]
mccL = [[] for _ in range(labelNum)]
mcMCCL = []
mcAccuracyL = []
confusionMatL = []
for testFileIDandClassifierID in testFileIDandClassifierIDs:
print('testFileIDandClassifierID = ' + str(testFileIDandClassifierID))
params_for_classifier = ParameterSetup(paramFileName='params.'+classifierID+'.json')
params_for_classifier.markovOrderForPrediction = markovOrder
(y_test, y_pred) = classifySequentially(params_for_classifier, paramID, paramDir, testFileIDandClassifierID)
print('y_test =', y_test)
print('type(y_test) =', type(y_test))
y_test = np.array(['W' if elem == 'RW' else elem for elem in y_test])
print('after replace: y_test =', y_test)
print('after replace: type(y_test) =', type(y_test))
# ignore ?'s in the beginning produced by
# i = 0
# while y_pred[i] == '?':
# i++
if params.classifierType == 'deep':
i = params.torch_lstm_length - 1 # remove from all clalssifiers because LSTM cannot predict first 9 elements.
else:
i = 0
print('for classifier ', testFileIDandClassifierID, ', first ', i, ' elements are removed.', sep='')
y_test, y_pred = y_test[i:], y_pred[i:]
(stageLabels, sensitivity, specificity, accuracy, precision, f1score) = y2sensitivity(y_test, y_pred)
(stageLabels4confusionMat, confusionMat) = y2confusionMat(y_test, y_pred, params.stageLabels4evaluation)
printConfusionMat(stageLabels4confusionMat, confusionMat)
# totalConfusionMat = totalConfusionMat + confusionMat
# print('y_test = ' + str(y_test[:50]))
# print('y_pred = ' + str(y_pred[:50]))
y_length = y_pred.shape[0]
print('stageLabels =', stageLabels)
print('labelNum = ' + str(labelNum))
for labelID in range(labelNum):
targetLabel = stageLabels[labelID]
sensitivityL[labelID].append(sensitivity[labelID])
specificityL[labelID].append(specificity[labelID])
accuracyL[labelID].append(accuracy[labelID])
precisionL[labelID].append(precision[labelID])
f1scoreL[labelID].append(f1score[labelID])
mcc = mathewsCorrelationCoefficient(stageLabels4confusionMat, confusionMat, targetLabel)
mccL[labelID].append(mcc)
print(' targetLabel = ' + targetLabel + ', sensitivity = ' + "{0:.3f}".format(sensitivity[labelID]) + ', specificity = ' + "{0:.3f}".format(specificity[labelID]) + ', accuracy = ' + "{0:.3f}".format(accuracy[labelID])+ ', precision = ' + "{0:.3f}".format(precision[labelID]))
print(' mcc for ' + targetLabel + ' = ' + "{0:.5f}".format(mcc))
mcMCCL.append(multiClassMCC(confusionMat))
print(' multi-class mcc = ' + "{0:.5f}".format(mcMCCL[-1]))
mcAccuracyL.append(sum(y_test == y_pred) / len(y_test))
print(' multi-class accuracy = ' + "{0:.5f}".format(mcAccuracyL[-1]))
confusionMatL.append(confusionMat)
print('')
writePredictionResults(testFileIDandClassifierID, params, y_test, y_pred, resultFileDescription)
if datasetType == 'test':
f = open(paramDir + '/test_result.' + classifierID + '.test.pkl', 'wb')
else:
f = open(paramDir + '/test_result.' + classifierID + '.pkl', 'wb')
pickle.dump((testFileIDandClassifierIDs, sensitivityL, specificityL, accuracyL, precisionL, f1scoreL, mccL, mcMCCL, mcAccuracyL, confusionMatL, stageLabels, fileNum, labelNum), f)
f.close()
#-----
# show the summary (average) of the result
print('Summary for classifierID ' + classifierID + ':')
printMetadata(params)
saveStatistics(params.pickledDir, classifierID, testFileIDandClassifierIDs, sensitivityL, specificityL, accuracyL, precisionL, f1scoreL, mccL, mcMCCL, mcAccuracyL, confusionMatL, stageLabels, fileNum, labelNum, datasetType)
# print('ch2TimeFrameNum = ' + str(params.ch2TimeFrameNum))
# print('binWidth4freqHisto = ' + str(params.binWidth4freqHisto))
sensitivityMeans, specificityMeans, accuracyMeans, precisionMean, f1scoreMean, mccMeans, mcMCCMean, mcAccuracyMean = meanStatistics(sensitivityL, specificityL, accuracyL, precisionL, f1scoreL, mccL, mcMCCL, mcAccuracyL, stageLabels, labelNum, fileNum)
# sensitivity_by_classifier_L.append(sensitivityMeans)
# specificity_by_classifier_L.append(specificityMeans)
# accuracy_by_classifier_L.append(accuracyMeans)
# precision_by_classifier_L.append(precisionMean)
# measures_by_classifier_L.append([sensitivityMeans, specificityMeans, accuracyMeans, precisionMean, f1scoreMean, mccMeans, mcMCCMean, mcAccuracyMean])
return [sensitivityMeans, specificityMeans, accuracyMeans, precisionMean, f1scoreMean, mccMeans, mcMCCMean, mcAccuracyMean]