gensim word2vecのモデルファイル読み込みで_pickle.UnpicklingErrorが出たときの対処法

python3.5.2

calc_w2v.py

import better_exceptions
from gensim.models import word2vec

file_name = "./model.txt"
model = word2vec.Word2Vec.load(file_name)
...

上記pythonコードで./model.txtというword2vecモデルファイル(binary=False)を読み込もうとすると、以下のようなエラーが出る。(bettter_exceptionsあり)

Traceback (most recent call last):
  File "calc_w2v.py", line 7, in <module>
    model = word2vec.Word2Vec.load(file_name)
            │                      └ './model.txt'
            └ <module 'gensim.models.word2vec' from '/Users/user/.pyenv/versions/anaconda3-4.2.0/lib/python3.5/site-packages/gensim/models/wo...
  File "/Users/user/.pyenv/versions/anaconda3-4.2.0/lib/python3.5/site-packages/gensim/models/word2vec.py", line 1520, in load
    model = super(Word2Vec, cls).load(*args, **kwargs)
                  │         │          │       └ {}
                  │         │          └ ('./model.txt',)
                  │         └ <class 'gensim.models.word2vec.Word2Vec'>
                  └ <class 'gensim.models.word2vec.Word2Vec'>
  File "/Users/user/.pyenv/versions/anaconda3-4.2.0/lib/python3.5/site-packages/gensim/utils.py", line 280, in load
    obj = unpickle(fname)
          │        └ './model.txt'
          └ <function unpickle at 0x10580e8c8>
  File "/Users/user/.pyenv/versions/anaconda3-4.2.0/lib/python3.5/site-packages/gensim/utils.py", line 933, in unpickle
    return _pickle.load(f, encoding='latin1')
           │            └ <_io.BufferedReader name='./model.txt'>
           └ <module 'pickle' from '/Users/user/.pyenv/versions/anaconda3-4.2.0/lib/python3.5/pickle.py'>
_pickle.UnpicklingError: could not find MARK

gensimで生成したファイルを読み込む際、load関数に対応するフォーマットとは違うことからエラーが出るらしい。読み込む関数を変える必要があるようだ。(詳しくはwebで。)

そこで、下記のようにpython3コードを修正。

calc_w2v.py

import better_exceptions
from gensim.models import word2vec

file_name = "./model.txt"
model = word2vec.KeyedVectors.load_word2vec_format(file_name)
...

無事に読み込めた。

---

calc_w2v.py

import better_exceptions
from gensim.models import word2vec
import sys
import cython

file_name = "./model.txt"
# model = word2vec.Word2Vec.load(file_name) ### _pickle.UnpicklingError: could not find MARK
model = word2vec.KeyedVectors.load_word2vec_format(file_name)

def word_subtraction(pos, neg):
    pos_list = [pos]
    neg_list = [neg]
    result = model.most_similar(positive=pos_list, negative=neg_list, topn = 10)
    for r in result:
        print(r)

pos = sys.argv[1]
neg = sys.argv[2]
print(model[pos])
print(model[neg])

print(pos, " - ", neg, " = ")
word_subtraction(pos, neg)

$ python3 calc_w2v.py アメリカ 日本