overall_segmented_accuracy.py

import os
import gettrace
import segment
import itertools
import inkml_to_pixels as itp
import pickle 

relative_path2011 = 'ICFHR_package/CROHME2011_data/CROHME_training/CROHME_training/'
relative_path2012 = 'ICFHR_package/CROHME2012_data/trainData/trainData/'

relative_testpath2011 = 'ICFHR_package/CROHME2011_data/CROHME_testGT/CROHME_testGT/'
relative_testpath2012 = 'ICFHR_package/CROHME2012_data/testDataGT/'

def getSegmentationAccuracy(pathArray):
  f = open('svm18px', "r")
  svm = pickle.load(f)
  
  j = 0

  total = 0
  total_correct = 0
  off_by_one = 0
  off_by_two = 0
  
  for path in pathArray:
    for f in os.listdir(path):
      j+=1
      if f[-5:] != "inkml": continue
      # print f
      traceList, symbolsList = gettrace.parseINKMLFile(path + f)
      segmentIndices = segment.segmentSymbols(traceList)

      correct = 0

      for label, elem in symbolsList:
        
        # Correctly classified
        if elem in segmentIndices:
          strokes = [traceList[i] for i in elem]
          pixels = itp.inkml_to_pixels(strokes)
          chain = list(itertools.chain(*pixels))
          chain.append(len(strokes))
          if svm.predict(chain)[0] == label:
            correct +=1
      if correct == len(symbolsList):
        total_correct +=1
      if correct + 1 >= len(symbolsList):
        off_by_one +=1
      if correct + 2 >= len(symbolsList):
        off_by_two +=1

          # cr.display(pixels)
      print j
  print total_correct
  print off_by_one
  print off_by_two

  print float(total_correct)/836
  print float(off_by_one)/836
  print float(off_by_two)/836

if __name__ == "__main__":
  # getSegmentationAccuracy([relative_path2011, relative_path2012])
  getSegmentationAccuracy([relative_testpath2011, relative_testpath2012])
  # getSegmentationAccuracy([relative_testpath2012])