def eval(self):
allValues = []
for dirPartition in self.dirPartitions:
#print "dirPartition", dirPartition
(annotateDir, testDir, trainDir, modelDir, resultDir) = self.partitions.getDirTestNames(dirPartition)
labeledContent = self._getFile(resultDir, 'testEstCRF_Wapiti.txt')
desiredContent = self._getFile(trainDir, 'evaldata_CRF_Wapiti.txt') # tmpFiles from training of testDir are saved in trainDir !
# harmonize the two lists, they are not tokenized the same way
desiredContentHarmonized, labeledContentHarmonized = prepareEval.prepareEval(desiredContent, labeledContent)
self._saveFile(labeledContentHarmonized, resultDir, 'annotatedEval.txt')
self._saveFile(desiredContentHarmonized, resultDir, 'desiredEval.txt')
evalText, labels, values = TokenAccuracyEval.evaluate(labeledContentHarmonized, desiredContentHarmonized)
allValues.append(values)
self._saveFile(evalText, dirPartition, 'evaluation.txt')
# calculate average of results for all partitions
average = [float(sum(col))/len(col) for col in zip(*allValues)]
allValues.append(average)
# print all results and average on the last line
finalEval = "\t".join(labels) + "\n"
finalEval += "\n".join(["\t".join(['{:f}'.format(v) for v in values]) for values in allValues])
self._saveFile(finalEval, self.partitions.getDirPercentName(), 'evaluation.tsv')
def tokenEval(self):
desiredContent = self._getFile(self.dirResult, 'evaldata_CRF_Wapiti.txt')
labeledContent= self._getFile(self.dirResult, 'testEstCRF_Wapiti.txt')
desiredContentHarmonized, labeledContentHarmonized = prepareEval.prepareEval(desiredContent, labeledContent)
evalText, labels, values = TokenAccuracyEval.evaluate(labeledContentHarmonized, desiredContentHarmonized)
finalEval = "\t".join(labels) + "\n"
finalEval += "\t".join(['{:f}'.format(v) for v in values]) + "\n"
finalEval += evalText
self._saveFile(finalEval, self.dirResult, 'evaluation.txt')