word prediction and ground truth alignment

align.py

from collections import Counter, defaultdict
from textdistance import levenshtein as lev
import numpy as np
import pdb
from tqdm import *
def CharMajVoting(words):

    def most_frequent(list_):
        counter = Counter(list_)
        return counter.most_common()[0][0]
    dict_ = defaultdict(list)
    lengths = [len(word) for word in words]
    common_length = most_frequent(lengths)
    for word in words:
        for i in range(len(word)):
            dict_[i].append(word[i])

    str_=''
    for i in range(len(dict_)):
        str_+=most_frequent(dict_[i])
    return str_[:common_length]

def similarity(word1, word2):
    return lev.normalized_distance(word1, word2)

def text_align(prWords, gtWords):
    row, col = len(prWords), len(gtWords)
    adjMat= np.zeros((row, col), dtype=float)
    for i in trange(len(prWords)):
        for j in range(len(gtWords)):
            adjMat[i, j] = similarity(prWords[i], gtWords[j])
    pr_aligned=[]
    for i in range(len(prWords)):
        nn = list(map(lambda x:gtWords[x], np.argsort(adjMat[i, :])[:1])) 
        pr_aligned.append((prWords[i], nn[0]))
    return pr_aligned