我如何通过消除错误来训练管道中的GaussianNB [AttributeError：'numpy.ndarray'对象没有属性'lower']

Question

This is the data i use count vectorizer and tfidftransformer and also use GaussianNB but i get error in this code. Please let me know the correct syntax.

train = [('I love this sandwich.','pos'),
     ('This is an amazing place!', 'pos'),
     ('I feel very good about these beers.', 'pos'),
     ('This is my best work.', 'pos'),
     ('What an awesome view', 'pos'),
     ('I do not like this restaurant', 'neg'),
     ('I am tired of this stuff.', 'neg'),
     ("I can't deal with this.", 'neg'),
     ('He is my sworn enemy!.', 'neg'),
     ('My boss is horrible.', 'neg')
    ]
from sklearn.feature_extraction.text import CountVectorizer
cv = CountVectorizer()

text_train_cv = cv.fit_transform(list(zip(*train))[0])
print(text_train_cv.toarray())

from sklearn.feature_extraction.text import TfidfTransformer
tfidf_trans = TfidfTransformer()

text_train_tfidf = tfidf_trans.fit_transform(text_train_cv)
print(text_train_tfidf.toarray())

from sklearn.naive_bayes import GaussianNB
clf = GaussianNB().fit(text_train_tfidf.toarray(), list(zip(*train))[1])

text_clf = Pipeline([('vect',CountVectorizer(stop_words='english')), 
('tfidf',TfidfTransformer()),('clf',GaussianNB(priors=None))])
text_clf = text_clf.fit(text_train_tfidf.toarray() , list(zip(*train))[1])
print(text_clf)

It give me error: AttributeError: 'numpy.ndarray' object has no attribute 'lower'

Answer 1

Do

clf = GaussianNB().fit(text_train_tfidf.toarray() , list(zip(*train))[1])

The GaussianNB doesnt support sparse matrices as input for X, but the TfidfTransformer will by default return a sparse matrix. Hence the error.

toarray() will convert that to dense. But note that it will lead to a high increase in memory usage.

Update:

When using a pipeline, you need to supply the data which you passed to transformer in the pipeline. In this case that is list(zip(*train))[0] .

text_clf = text_clf.fit(list(zip(*train))[0] , list(zip(*train))[1])

That will solve your first error. But you will still get an error because of sparse matrix. See this answer for solving that :- https://stackoverflow.com/a/28384887/3374996

Answer 2

MultinomialNB is used very often for text classification tasks and it does support sparse matrices as an input data set.

PS using dense matrices for bigger corpuses you might end up with the MemoryError

So try this:

from sklearn.naive_bayes import MultinomialNB

clf = MultinomialNB().fit(text_train_tfidf , list(zip(*train))[1])