src/main.py

   1 from common_dmv import ROOTNUM
   2 from wsjdep import WSJDepCorpusReader
   3 from loc_h_dmv import DMV_Grammar, reestimate, DEBUG, mpp
   4 from loc_h_harmonic import initialize
   5
   6 #             try:
   7 #                 return add_root(parse)
   8 #             except:
   9 #                 print "No single possible root, todo what?"
  10 #                 return parse
  11 #todo
  12
  13 # todo:
  14 def evaluate(g, tagged_and_parsed_sents):
  15     '''
  16     tagged_and_parsed_sents is a list of pairs:
  17     (tagonly_sent, parsed_sent)
  18
  19     R_num += 1 if pair from parsed is in mpp
  20     R_den += 1 per pair from parsed
  21
  22     P_num += 1 if pair from mpp is in parsed
  23     P_den += 1 per pair from mpp
  24
  25     F1 = (2 * P * R)/(P + R), harmonisk snitt av P og R
  26     '''
  27     recall_num = 0
  28     recall_den = 0
  29     precision_num = 0
  30     precision_den = 0
  31
  32     for sent, parse in tagged_and_parsed_sents:
  33         mpp_sent = mpp(g, sent)
  34         for pair in parse:
  35             recall_den += 1
  36             if pair in mpp_sent: recall_num += 1
  37         for pair in mpp_sent:
  38             if pair[0][0] == ROOTNUM: continue # todo: add ROOT to parses?
  39             precision_den += 1
  40             if pair in parse: precision_num += 1
  41
  42     recall = recall_num / recall_den
  43     precision = precision_num / precision_den
  44     F1 = (2 * recall * precision) / (precision + recall)
  45
  46     return "Recall: %.4f\tPrecision: %.4f\tF1: %.4f"%(recall, precision, F1)
  47
  48
  49 if __name__ == "__main__":
  50     print "main.py:"
  51     reader = WSJDepCorpusReader(None)
  52
  53     tagonlys = reader.tagonly_sents()
  54     g = initialize(tagonlys[0:500])
  55     print g
  56
  57 #     DEBUG.add('REEST')
  58 #     f = reestimate(g, reader.tagonly_sents())
  59 #     print g
  60
  61     print evaluate(g, tags_and_parses[0:500])
  62
  63     print "main.py: done"