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 # todo:
   7 def evaluate(g, tagged_and_parsed_sents):
   8     '''
   9     tagged_and_parsed_sents is a list of pairs:
  10     (tagonly_sent, parsed_sent)
  11
  12     R_num += 1 if pair from parsed is in mpp
  13     R_den += 1 per pair from parsed
  14
  15     P_num += 1 if pair from mpp is in parsed
  16     P_den += 1 per pair from mpp
  17
  18     F1 = (2 * P * R)/(P + R), harmonisk snitt av P og R
  19     '''
  20     recall_num = 0
  21     recall_den = 0
  22     precision_num = 0
  23     precision_den = 0
  24
  25     for sent, parse in tagged_and_parsed_sents:
  26         mpp_sent = mpp(g, sent)
  27         for pair in parse:
  28             recall_den += 1
  29             if pair in mpp_sent: recall_num += 1
  30         for pair in mpp_sent:
  31             if pair[0][0] == ROOTNUM: continue # todo: add ROOT to parses?
  32             precision_den += 1
  33             if pair in parse: precision_num += 1
  34
  35     recall = recall_num / recall_den
  36     precision = precision_num / precision_den
  37     F1 = (2 * recall * precision) / (precision + recall)
  38
  39     return "Recall: %.4f\tPrecision: %.4f\tF1: %.4f"%(recall, precision, F1)
  40
  41
  42 if __name__ == "__main__":
  43     print "main.py:"
  44     reader = WSJDepCorpusReader(None)
  45
  46     tagonlys = reader.tagonly_sents()
  47     g = initialize(tagonlys[0:500])
  48     print g
  49
  50 #     DEBUG.add('REEST')
  51 #     f = reestimate(g, reader.tagonly_sents())
  52 #     print g
  53
  54     print evaluate(g, tags_and_parses[0:500])
  55
  56     print "main.py: done"