evaluation.py

from numpy import *
import pandas as pd
import Levenshtein
import re


def cleanASJP(word):
    """takes an ASJP string as argument
    and returns the string with all diacritics removed."""
    word = re.sub(r",","-",word)
    word = re.sub(r"\%","",word)
    word = re.sub(r"\*","",word)
    word = re.sub(r"\"","",word)
    word = re.sub(r".~","",word)
    word = re.sub(r"(.)(.)(.)\$",r"\2",word)
    word = re.sub(r"\$","",word)
    word = re.sub(r"\s+","",word)
    return word.replace('~','')

asjp = pd.read_table('dataset.tab',index_col=0,
                     sep='\t',na_filter=False)


romance = array([x for x in asjp[asjp.wls_gen=='ROMANCE'].index
                 if x!='LATIN'])




reconstruction = pd.read_csv('reconstruction.csv',index_col=0)

concepts = array(reconstruction.index)

def ldn(a,b):
    return min([1.*Levenshtein.distance(x,y)/max(len(x),len(y))
                for x in a.split('-') for y in b.split('-')])


romanceCleaned = pd.DataFrame([[cleanASJP(x).split('-')[0] for x in y]
                               for y in asjp.ix[romance][concepts].values],
                              index=romance,
                              columns=concepts)

latinCleaned = pd.Series([cleanASJP(x) for x in reconstruction.Latin.values],
                         index=reconstruction.index)[concepts]



reconEval = mean([ldn(x,y) for x,y in zip(reconstruction.reconstruction.values,
                                          latinCleaned.values)])

romanceEval = pd.Series([mean([ldn(romanceCleaned.ix[l][c],latinCleaned[c])
                               for c in concepts])
                         for l in romance],
                        index=romance)
romanceEval.ix['Proto-Romance'] = reconEval

romanceEval.to_csv('romanceEvaluation.csv')