如何为不同类别的scikit learn分类器获得最丰富的信息？

Question

NLTK包提供了一个方法show_most_informative_features()来查找这两个类的最重要特性，输出如下：

   contains(outstanding) = True              pos : neg    =     11.1 : 1.0
        contains(seagal) = True              neg : pos    =      7.7 : 1.0
   contains(wonderfully) = True              pos : neg    =      6.8 : 1.0
         contains(damon) = True              pos : neg    =      5.9 : 1.0
        contains(wasted) = True              neg : pos    =      5.8 : 1.0

正如在这个问题How to get most informative features for scikit-learn classifiers?中所回答的，这也可以在scikit learn中工作。然而，对于二进制分类器，这个问题的答案只输出最好的特征本身。

所以我的问题是，我如何识别特征的相关类，就像上面的例子一样（杰出的在pos类中信息量最大，而seagal在否定类中信息量最大）？

编辑：事实上，我想要的是每节课最具信息性的单词列表。我该怎么做？谢谢！

Answer 1

基本上你需要：

def most_informative_feature_for_class(vectorizer, classifier, classlabel, n=10):
    labelid = list(classifier.classes_).index(classlabel)
    feature_names = vectorizer.get_feature_names()
    topn = sorted(zip(classifier.coef_[labelid], feature_names))[-n:]

    for coef, feat in topn:
        print classlabel, feat, coef    

• classifier.classes_访问分类器中的类标签的索引

• vectorizer.get_feature_names()是不言而喻的

• sorted(zip(classifier.coef_[labelid], feature_names))[-n:]检索给定类标签的分类器系数，然后按升序对其排序。

---

我将使用https://github.com/alvations/bayesline中的一个简单示例

输入文件train.txt：

$ echo """Pošto je EULEX obećao da će obaviti istragu o prošlosedmičnom izbijanju nasilja na sjeveru Kosova, taj incident predstavlja još jedan ispit kapaciteta misije da doprinese jačanju vladavine prava.
> De todas as provações que teve de suplantar ao longo da vida, qual foi a mais difícil? O início. Qualquer começo apresenta dificuldades que parecem intransponíveis. Mas tive sempre a minha mãe do meu lado. Foi ela quem me ajudou a encontrar forças para enfrentar as situações mais decepcionantes, negativas, as que me punham mesmo furiosa.
> Al parecer, Andrea Guasch pone que una relación a distancia es muy difícil de llevar como excusa. Algo con lo que, por lo visto, Alex Lequio no está nada de acuerdo. ¿O es que más bien ya ha conseguido la fama que andaba buscando?
> Vo väčšine golfových rezortov ide o veľký komplex niekoľkých ihrísk blízko pri sebe spojených s hotelmi a ďalšími možnosťami trávenia voľného času – nie vždy sú manželky či deti nadšenými golfistami, a tak potrebujú iný druh vyžitia. Zaujímavé kombinácie ponúkajú aj rakúske, švajčiarske či talianske Alpy, kde sa dá v zime lyžovať a v lete hrať golf pod vysokými alpskými končiarmi.""" > test.in

代码：

import codecs, re, time
from itertools import chain

import numpy as np

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.naive_bayes import MultinomialNB

trainfile = 'train.txt'

# Vectorizing data.
train = []
word_vectorizer = CountVectorizer(analyzer='word')
trainset = word_vectorizer.fit_transform(codecs.open(trainfile,'r','utf8'))
tags = ['bs','pt','es','sr']

# Training NB
mnb = MultinomialNB()
mnb.fit(trainset, tags)

def most_informative_feature_for_class(vectorizer, classifier, classlabel, n=10):
    labelid = list(classifier.classes_).index(classlabel)
    feature_names = vectorizer.get_feature_names()
    topn = sorted(zip(classifier.coef_[labelid], feature_names))[-n:]

    for coef, feat in topn:
        print classlabel, feat, coef



most_informative_feature_for_class(word_vectorizer, mnb, 'bs')
print 
most_informative_feature_for_class(word_vectorizer, mnb, 'pt')

[出局]：

bs obećao -4.50534985071
bs pošto -4.50534985071
bs prava -4.50534985071
bs predstavlja -4.50534985071
bs prošlosedmičnom -4.50534985071
bs sjeveru -4.50534985071
bs taj -4.50534985071
bs vladavine -4.50534985071
bs će -4.50534985071
bs da -4.0998847426

pt teve -4.63472898823
pt tive -4.63472898823
pt todas -4.63472898823
pt vida -4.63472898823
pt de -4.22926388012
pt foi -4.22926388012
pt mais -4.22926388012
pt me -4.22926388012
pt as -3.94158180767
pt que -3.94158180767

Answer 2

左右两边的两个类可以得到相同的结果：

           precision    recall  f1-score   support

 Irrelevant       0.77      0.98      0.86       129
   Relevant       0.78      0.15      0.25        46

avg / total       0.77      0.77      0.70       175

    -1.3914 davis                   1.4809  austin
    -1.1023 suicide                 1.0695  march
    -1.0609 arrested                1.0379  call
    -1.0145 miller                  1.0152  tsa
    -0.8902 packers                 0.9848  passengers
    -0.8370 train                   0.9547  pensacola
    -0.7557 trevor                  0.7432  bag
    -0.7457 near                    0.7056  conditt
    -0.7359 military                0.7002  midamerica
    -0.7302 berlin                  0.6987  mark
    -0.6880 april                   0.6799  grenade
    -0.6581 plane                   0.6357  suspicious
    -0.6351 disposal                0.6348  death
    -0.5804 wwii                    0.6053  flight
    -0.5723 terminal                0.5745  marabi


def Show_most_informative_features(vectorizer, clf, n=20):
    feature_names = vectorizer.get_feature_names()
    coefs_with_fns = sorted(zip(clf.coef_[0], feature_names))
    top = zip(coefs_with_fns[:n], coefs_with_fns[:-(n + 1):-1])
    for (coef_1, fn_1), (coef_2, fn_2) in top:
      print ("\t%.4f\t%-15s\t\t%.4f\t%-15s" % (coef_1, fn_1, coef_2, fn_2))

Answer 3

在二值分类的情况下，系数数组似乎已经变平。

让我们尝试用两个标签重新标记数据：

import codecs, re, time
from itertools import chain

import numpy as np

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.naive_bayes import MultinomialNB

trainfile = 'train.txt'

# Vectorizing data.
train = []
word_vectorizer = CountVectorizer(analyzer='word')
trainset = word_vectorizer.fit_transform(codecs.open(trainfile,'r','utf8'))
tags = ['bs','pt','bs','pt']

# Training NB
mnb = MultinomialNB()
mnb.fit(trainset, tags)

print mnb.classes_
print mnb.coef_[0]
print mnb.coef_[1]

[出局]：

['bs' 'pt']
[-5.55682806 -4.86368088 -4.86368088 -5.55682806 -5.55682806 -5.55682806
 -4.86368088 -4.86368088 -5.55682806 -5.55682806 -4.86368088 -4.86368088
 -4.1705337  -5.55682806 -4.86368088 -5.55682806 -4.86368088 -5.55682806
 -5.55682806 -5.55682806 -4.86368088 -4.45821577 -4.86368088 -4.86368088
 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -5.55682806 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -5.55682806
 -5.55682806 -5.55682806 -5.55682806 -4.45821577 -4.86368088 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -5.55682806 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088
 -4.86368088 -5.55682806 -5.55682806 -5.55682806 -5.55682806 -5.55682806
 -5.55682806 -5.55682806 -5.55682806 -4.86368088 -4.86368088 -4.86368088
 -4.86368088 -5.55682806 -5.55682806 -4.86368088 -5.55682806 -4.86368088
 -5.55682806 -5.55682806 -4.86368088 -4.86368088 -4.45821577 -4.86368088
 -4.86368088 -4.45821577 -4.86368088 -4.86368088 -4.86368088 -5.55682806
 -4.86368088 -5.55682806 -5.55682806 -4.86368088 -5.55682806 -5.55682806
 -4.86368088 -5.55682806 -4.86368088 -4.86368088 -4.86368088 -5.55682806
 -5.55682806 -5.55682806 -4.86368088 -4.86368088 -5.55682806 -4.86368088
 -5.55682806 -4.86368088 -5.55682806 -4.86368088 -5.55682806 -5.55682806
 -5.55682806 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088
 -4.86368088 -4.1705337  -4.86368088 -4.86368088 -5.55682806 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -4.86368088
 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088
 -5.55682806 -5.55682806 -4.86368088 -4.86368088 -4.86368088 -4.86368088
 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088 -5.55682806
 -4.86368088 -4.45821577 -4.86368088 -4.86368088]
Traceback (most recent call last):
  File "test.py", line 24, in <module>
    print mnb.coef_[1]
IndexError: index 1 is out of bounds for axis 0 with size 1

所以让我们做一些诊断：

print mnb.feature_count_
print mnb.coef_[0]

[出局]：

[[ 1.  0.  0.  1.  1.  1.  0.  0.  1.  1.  0.  0.  0.  1.  0.  1.  0.  1.
   1.  1.  2.  2.  0.  0.  0.  1.  1.  0.  1.  0.  0.  0.  0.  0.  2.  1.
   1.  1.  1.  0.  0.  0.  0.  0.  0.  1.  1.  0.  0.  0.  0.  1.  0.  0.
   0.  1.  1.  1.  1.  1.  1.  1.  1.  0.  0.  0.  0.  1.  1.  0.  1.  0.
   1.  2.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  1.  1.  0.  1.  1.
   0.  1.  0.  0.  0.  1.  1.  1.  0.  0.  1.  0.  1.  0.  1.  0.  1.  1.
   1.  0.  0.  1.  0.  0.  0.  4.  0.  0.  1.  0.  0.  0.  0.  0.  1.  0.
   0.  0.  1.  0.  0.  0.  0.  0.  0.  1.  0.  0.  1.  1.  0.  0.  0.  0.
   0.  0.  1.  0.  0.  1.  0.  0.  0.  0.]
 [ 0.  1.  1.  0.  0.  0.  1.  1.  0.  0.  1.  1.  3.  0.  1.  0.  1.  0.
   0.  0.  1.  2.  1.  1.  1.  1.  0.  1.  0.  1.  1.  1.  1.  1.  0.  0.
   0.  0.  0.  2.  1.  1.  1.  1.  1.  0.  0.  1.  1.  1.  1.  0.  1.  1.
   1.  0.  0.  0.  0.  0.  0.  0.  0.  1.  1.  1.  1.  0.  0.  1.  0.  1.
   0.  0.  1.  1.  2.  1.  1.  2.  1.  1.  1.  0.  1.  0.  0.  1.  0.  0.
   1.  0.  1.  1.  1.  0.  0.  0.  1.  1.  0.  1.  0.  1.  0.  1.  0.  0.
   0.  1.  1.  0.  1.  1.  1.  3.  1.  1.  0.  1.  1.  1.  1.  1.  0.  1.
   1.  1.  0.  1.  1.  1.  1.  1.  1.  0.  1.  1.  0.  0.  1.  1.  1.  1.
   1.  1.  0.  1.  1.  0.  1.  2.  1.  1.]]
[-5.55682806 -4.86368088 -4.86368088 -5.55682806 -5.55682806 -5.55682806
 -4.86368088 -4.86368088 -5.55682806 -5.55682806 -4.86368088 -4.86368088
 -4.1705337  -5.55682806 -4.86368088 -5.55682806 -4.86368088 -5.55682806
 -5.55682806 -5.55682806 -4.86368088 -4.45821577 -4.86368088 -4.86368088
 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -5.55682806 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -5.55682806
 -5.55682806 -5.55682806 -5.55682806 -4.45821577 -4.86368088 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -5.55682806 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088
 -4.86368088 -5.55682806 -5.55682806 -5.55682806 -5.55682806 -5.55682806
 -5.55682806 -5.55682806 -5.55682806 -4.86368088 -4.86368088 -4.86368088
 -4.86368088 -5.55682806 -5.55682806 -4.86368088 -5.55682806 -4.86368088
 -5.55682806 -5.55682806 -4.86368088 -4.86368088 -4.45821577 -4.86368088
 -4.86368088 -4.45821577 -4.86368088 -4.86368088 -4.86368088 -5.55682806
 -4.86368088 -5.55682806 -5.55682806 -4.86368088 -5.55682806 -5.55682806
 -4.86368088 -5.55682806 -4.86368088 -4.86368088 -4.86368088 -5.55682806
 -5.55682806 -5.55682806 -4.86368088 -4.86368088 -5.55682806 -4.86368088
 -5.55682806 -4.86368088 -5.55682806 -4.86368088 -5.55682806 -5.55682806
 -5.55682806 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088
 -4.86368088 -4.1705337  -4.86368088 -4.86368088 -5.55682806 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -4.86368088
 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088 -4.86368088
 -4.86368088 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088
 -5.55682806 -5.55682806 -4.86368088 -4.86368088 -4.86368088 -4.86368088
 -4.86368088 -4.86368088 -5.55682806 -4.86368088 -4.86368088 -5.55682806
 -4.86368088 -4.45821577 -4.86368088 -4.86368088]

似乎这些特征都被计算在内，然后当矢量化时，为了节省内存，它被展平，所以让我们尝试：

index = 0
coef_features_c1_c2 = []

for feat, c1, c2 in zip(word_vectorizer.get_feature_names(), mnb.feature_count_[0], mnb.feature_count_[1]):
    coef_features_c1_c2.append(tuple([mnb.coef_[0][index], feat, c1, c2]))
    index+=1

for i in sorted(coef_features_c1_c2):
    print i

[出局]：

(-5.5568280616995374, u'acuerdo', 1.0, 0.0)
(-5.5568280616995374, u'al', 1.0, 0.0)
(-5.5568280616995374, u'alex', 1.0, 0.0)
(-5.5568280616995374, u'algo', 1.0, 0.0)
(-5.5568280616995374, u'andaba', 1.0, 0.0)
(-5.5568280616995374, u'andrea', 1.0, 0.0)
(-5.5568280616995374, u'bien', 1.0, 0.0)
(-5.5568280616995374, u'buscando', 1.0, 0.0)
(-5.5568280616995374, u'como', 1.0, 0.0)
(-5.5568280616995374, u'con', 1.0, 0.0)
(-5.5568280616995374, u'conseguido', 1.0, 0.0)
(-5.5568280616995374, u'distancia', 1.0, 0.0)
(-5.5568280616995374, u'doprinese', 1.0, 0.0)
(-5.5568280616995374, u'es', 2.0, 0.0)
(-5.5568280616995374, u'est\xe1', 1.0, 0.0)
(-5.5568280616995374, u'eulex', 1.0, 0.0)
(-5.5568280616995374, u'excusa', 1.0, 0.0)
(-5.5568280616995374, u'fama', 1.0, 0.0)
(-5.5568280616995374, u'guasch', 1.0, 0.0)
(-5.5568280616995374, u'ha', 1.0, 0.0)
(-5.5568280616995374, u'incident', 1.0, 0.0)
(-5.5568280616995374, u'ispit', 1.0, 0.0)
(-5.5568280616995374, u'istragu', 1.0, 0.0)
(-5.5568280616995374, u'izbijanju', 1.0, 0.0)
(-5.5568280616995374, u'ja\u010danju', 1.0, 0.0)
(-5.5568280616995374, u'je', 1.0, 0.0)
(-5.5568280616995374, u'jedan', 1.0, 0.0)
(-5.5568280616995374, u'jo\u0161', 1.0, 0.0)
(-5.5568280616995374, u'kapaciteta', 1.0, 0.0)
(-5.5568280616995374, u'kosova', 1.0, 0.0)
(-5.5568280616995374, u'la', 1.0, 0.0)
(-5.5568280616995374, u'lequio', 1.0, 0.0)
(-5.5568280616995374, u'llevar', 1.0, 0.0)
(-5.5568280616995374, u'lo', 2.0, 0.0)
(-5.5568280616995374, u'misije', 1.0, 0.0)
(-5.5568280616995374, u'muy', 1.0, 0.0)
(-5.5568280616995374, u'm\xe1s', 1.0, 0.0)
(-5.5568280616995374, u'na', 1.0, 0.0)
(-5.5568280616995374, u'nada', 1.0, 0.0)
(-5.5568280616995374, u'nasilja', 1.0, 0.0)
(-5.5568280616995374, u'no', 1.0, 0.0)
(-5.5568280616995374, u'obaviti', 1.0, 0.0)
(-5.5568280616995374, u'obe\u0107ao', 1.0, 0.0)
(-5.5568280616995374, u'parecer', 1.0, 0.0)
(-5.5568280616995374, u'pone', 1.0, 0.0)
(-5.5568280616995374, u'por', 1.0, 0.0)
(-5.5568280616995374, u'po\u0161to', 1.0, 0.0)
(-5.5568280616995374, u'prava', 1.0, 0.0)
(-5.5568280616995374, u'predstavlja', 1.0, 0.0)
(-5.5568280616995374, u'pro\u0161losedmi\u010dnom', 1.0, 0.0)
(-5.5568280616995374, u'relaci\xf3n', 1.0, 0.0)
(-5.5568280616995374, u'sjeveru', 1.0, 0.0)
(-5.5568280616995374, u'taj', 1.0, 0.0)
(-5.5568280616995374, u'una', 1.0, 0.0)
(-5.5568280616995374, u'visto', 1.0, 0.0)
(-5.5568280616995374, u'vladavine', 1.0, 0.0)
(-5.5568280616995374, u'ya', 1.0, 0.0)
(-5.5568280616995374, u'\u0107e', 1.0, 0.0)
(-4.863680881139592, u'aj', 0.0, 1.0)
(-4.863680881139592, u'ajudou', 0.0, 1.0)
(-4.863680881139592, u'alpsk\xfdmi', 0.0, 1.0)
(-4.863680881139592, u'alpy', 0.0, 1.0)
(-4.863680881139592, u'ao', 0.0, 1.0)
(-4.863680881139592, u'apresenta', 0.0, 1.0)
(-4.863680881139592, u'bl\xedzko', 0.0, 1.0)
(-4.863680881139592, u'come\xe7o', 0.0, 1.0)
(-4.863680881139592, u'da', 2.0, 1.0)
(-4.863680881139592, u'decepcionantes', 0.0, 1.0)
(-4.863680881139592, u'deti', 0.0, 1.0)
(-4.863680881139592, u'dificuldades', 0.0, 1.0)
(-4.863680881139592, u'dif\xedcil', 1.0, 1.0)
(-4.863680881139592, u'do', 0.0, 1.0)
(-4.863680881139592, u'druh', 0.0, 1.0)
(-4.863680881139592, u'd\xe1', 0.0, 1.0)
(-4.863680881139592, u'ela', 0.0, 1.0)
(-4.863680881139592, u'encontrar', 0.0, 1.0)
(-4.863680881139592, u'enfrentar', 0.0, 1.0)
(-4.863680881139592, u'for\xe7as', 0.0, 1.0)
(-4.863680881139592, u'furiosa', 0.0, 1.0)
(-4.863680881139592, u'golf', 0.0, 1.0)
(-4.863680881139592, u'golfistami', 0.0, 1.0)
(-4.863680881139592, u'golfov\xfdch', 0.0, 1.0)
(-4.863680881139592, u'hotelmi', 0.0, 1.0)
(-4.863680881139592, u'hra\u0165', 0.0, 1.0)
(-4.863680881139592, u'ide', 0.0, 1.0)
(-4.863680881139592, u'ihr\xedsk', 0.0, 1.0)
(-4.863680881139592, u'intranspon\xedveis', 0.0, 1.0)
(-4.863680881139592, u'in\xedcio', 0.0, 1.0)
(-4.863680881139592, u'in\xfd', 0.0, 1.0)
(-4.863680881139592, u'kde', 0.0, 1.0)
(-4.863680881139592, u'kombin\xe1cie', 0.0, 1.0)
(-4.863680881139592, u'komplex', 0.0, 1.0)
(-4.863680881139592, u'kon\u010diarmi', 0.0, 1.0)
(-4.863680881139592, u'lado', 0.0, 1.0)
(-4.863680881139592, u'lete', 0.0, 1.0)
(-4.863680881139592, u'longo', 0.0, 1.0)
(-4.863680881139592, u'ly\u017eova\u0165', 0.0, 1.0)
(-4.863680881139592, u'man\u017eelky', 0.0, 1.0)
(-4.863680881139592, u'mas', 0.0, 1.0)
(-4.863680881139592, u'mesmo', 0.0, 1.0)
(-4.863680881139592, u'meu', 0.0, 1.0)
(-4.863680881139592, u'minha', 0.0, 1.0)
(-4.863680881139592, u'mo\u017enos\u0165ami', 0.0, 1.0)
(-4.863680881139592, u'm\xe3e', 0.0, 1.0)
(-4.863680881139592, u'nad\u0161en\xfdmi', 0.0, 1.0)
(-4.863680881139592, u'negativas', 0.0, 1.0)
(-4.863680881139592, u'nie', 0.0, 1.0)
(-4.863680881139592, u'nieko\u013ek\xfdch', 0.0, 1.0)
(-4.863680881139592, u'para', 0.0, 1.0)
(-4.863680881139592, u'parecem', 0.0, 1.0)
(-4.863680881139592, u'pod', 0.0, 1.0)
(-4.863680881139592, u'pon\xfakaj\xfa', 0.0, 1.0)
(-4.863680881139592, u'potrebuj\xfa', 0.0, 1.0)
(-4.863680881139592, u'pri', 0.0, 1.0)
(-4.863680881139592, u'prova\xe7\xf5es', 0.0, 1.0)
(-4.863680881139592, u'punham', 0.0, 1.0)
(-4.863680881139592, u'qual', 0.0, 1.0)
(-4.863680881139592, u'qualquer', 0.0, 1.0)
(-4.863680881139592, u'quem', 0.0, 1.0)
(-4.863680881139592, u'rak\xfaske', 0.0, 1.0)
(-4.863680881139592, u'rezortov', 0.0, 1.0)
(-4.863680881139592, u'sa', 0.0, 1.0)
(-4.863680881139592, u'sebe', 0.0, 1.0)
(-4.863680881139592, u'sempre', 0.0, 1.0)
(-4.863680881139592, u'situa\xe7\xf5es', 0.0, 1.0)
(-4.863680881139592, u'spojen\xfdch', 0.0, 1.0)
(-4.863680881139592, u'suplantar', 0.0, 1.0)
(-4.863680881139592, u's\xfa', 0.0, 1.0)
(-4.863680881139592, u'tak', 0.0, 1.0)
(-4.863680881139592, u'talianske', 0.0, 1.0)
(-4.863680881139592, u'teve', 0.0, 1.0)
(-4.863680881139592, u'tive', 0.0, 1.0)
(-4.863680881139592, u'todas', 0.0, 1.0)
(-4.863680881139592, u'tr\xe1venia', 0.0, 1.0)
(-4.863680881139592, u've\u013ek\xfd', 0.0, 1.0)
(-4.863680881139592, u'vida', 0.0, 1.0)
(-4.863680881139592, u'vo', 0.0, 1.0)
(-4.863680881139592, u'vo\u013en\xe9ho', 0.0, 1.0)
(-4.863680881139592, u'vysok\xfdmi', 0.0, 1.0)
(-4.863680881139592, u'vy\u017eitia', 0.0, 1.0)
(-4.863680881139592, u'v\xe4\u010d\u0161ine', 0.0, 1.0)
(-4.863680881139592, u'v\u017edy', 0.0, 1.0)
(-4.863680881139592, u'zauj\xedmav\xe9', 0.0, 1.0)
(-4.863680881139592, u'zime', 0.0, 1.0)
(-4.863680881139592, u'\u010dasu', 0.0, 1.0)
(-4.863680881139592, u'\u010fal\u0161\xedmi', 0.0, 1.0)
(-4.863680881139592, u'\u0161vaj\u010diarske', 0.0, 1.0)
(-4.4582157730314274, u'de', 2.0, 2.0)
(-4.4582157730314274, u'foi', 0.0, 2.0)
(-4.4582157730314274, u'mais', 0.0, 2.0)
(-4.4582157730314274, u'me', 0.0, 2.0)
(-4.4582157730314274, u'\u010di', 0.0, 2.0)
(-4.1705337005796466, u'as', 0.0, 3.0)
(-4.1705337005796466, u'que', 4.0, 3.0)

现在我们看到一些模式。。。似乎较高的系数有利于一个类，而另一个尾部有利于另一个类，所以您可以简单地执行以下操作：

import codecs, re, time
from itertools import chain

import numpy as np

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.naive_bayes import MultinomialNB

trainfile = 'train.txt'

# Vectorizing data.
train = []
word_vectorizer = CountVectorizer(analyzer='word')
trainset = word_vectorizer.fit_transform(codecs.open(trainfile,'r','utf8'))
tags = ['bs','pt','bs','pt']

# Training NB
mnb = MultinomialNB()
mnb.fit(trainset, tags)

def most_informative_feature_for_binary_classification(vectorizer, classifier, n=10):
    class_labels = classifier.classes_
    feature_names = vectorizer.get_feature_names()
    topn_class1 = sorted(zip(classifier.coef_[0], feature_names))[:n]
    topn_class2 = sorted(zip(classifier.coef_[0], feature_names))[-n:]

    for coef, feat in topn_class1:
        print class_labels[0], coef, feat

    print

    for coef, feat in reversed(topn_class2):
        print class_labels[1], coef, feat


most_informative_feature_for_binary_classification(word_vectorizer, mnb)

[出局]：

bs -5.5568280617 acuerdo
bs -5.5568280617 al
bs -5.5568280617 alex
bs -5.5568280617 algo
bs -5.5568280617 andaba
bs -5.5568280617 andrea
bs -5.5568280617 bien
bs -5.5568280617 buscando
bs -5.5568280617 como
bs -5.5568280617 con

pt -4.17053370058 que
pt -4.17053370058 as
pt -4.45821577303 či
pt -4.45821577303 me
pt -4.45821577303 mais
pt -4.45821577303 foi
pt -4.45821577303 de
pt -4.86368088114 švajčiarske
pt -4.86368088114 ďalšími
pt -4.86368088114 času

实际上，如果您仔细阅读了@larsmans注释，他在How to get most informative features for scikit-learn classifiers?中给出了关于二进制类系数的提示