How to classify text pairs using scikit-learn?

Question

I have read many different blogs on this topic, but haven't been able to find a clear solution. I have the following scenario:

I have a list of pairs of texts with labels 1, or -1.
For each text pair , I want the features to be a concatenation in the following fashion: f () = tfidf(t1) "concat" tfidf(t2)

Any suggestions on how to do the same ? I have the following code but it gives an error:

    count_vect = TfidfVectorizer(analyzer=u'char', ngram_range=ngram_range)
    X0_train_counts = count_vect.fit_transform([x[0] for x in training_documents])
    X1_train_counts = count_vect.fit_transform([x[1] for x in training_documents])
    combined_features = FeatureUnion([("x0", X0_train_counts), ("x1", X1_train_counts)])
    clf = LinearSVC().fit(combined_features, training_target)
    average_training_accuracy += clf.score(combined_features, training_target)

Here's the error I get:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
scoreEdgesUsingClassifier(None, pos, neg, 1,ngram_range=(2,5), max_size=1000000, test_size=100000)

 scoreEdgesUsingClassifier(unc, pos, neg, number_of_iterations, ngram_range, max_size, test_size)
 X0_train_counts = count_vect.fit_transform([x[0] for x in training_documents])
 X1_train_counts = count_vect.fit_transform([x[1] for x in training_documents])
 combined_features = FeatureUnion([("x0", X0_train_counts), ("x1", X1_train_counts)])
 print "Done transforming, now training classifier"

lib/python2.7/site-packages/sklearn/pipeline.pyc in __init__(self, transformer_list, n_jobs, transformer_weights)
616         self.n_jobs = n_jobs
617         self.transformer_weights = transformer_weights
--> 618         self._validate_transformers()
619 
620     def get_params(self, deep=True):

lib/python2.7/site-packages/sklearn/pipeline.pyc in _validate_transformers(self)
660                 raise TypeError("All estimators should implement fit and "
661                                 "transform. '%s' (type %s) doesn't" %
--> 662                                 (t, type(t)))
663 
664     def _iter(self):

TypeError: All estimators should implement fit and transform. '  (0, 49025) 0.0575144797079

 (254741, 38401)    0.184394443164
 (254741, 201747)   0.186080393768
 (254741, 179231)   0.195062580945
 (254741, 156925)   0.211367771299
 (254741, 90026)    0.202458920022' (type <class 'scipy.sparse.csr.csr_matrix'>) doesn't

Update

Here's the solution:

    count_vect = TfidfVectorizer(analyzer=u'char', ngram_range=ngram_range)
    training_docs_combined = [x[0] for x in training_documents] + [x[1] for x in training_documents]        
    X_train_counts = count_vect.fit_transform(training_docs_combined)
    concat_features  = hstack((X_train_counts[0:len(training_docs_combined) / 2 ], X_train_counts[len (training_docs_combined) / 2:]))

    clf = LinearSVC().fit(concat_features, training_target)
    average_training_accuracy += clf.score(concat_features, training_target)

The labels are for a pair of texts, not a single text? What error are you getting? — rth
– rth, Commented Mar 16, 2017 at 20:51

rth · Accepted Answer · 2017-03-16 21:04:01Z

2

FeatureUnion from scikit-learn takes as input estimators, not data arrays.

You can either concatenate the resulting X0_train_counts, X1_train_counts arrays simply with scipy.sparse.hstack, or create two independent instances of TfidfVectorizer, apply FeatureUnion to them, and then call the fit_transform method.

answered Mar 16, 2017 at 21:04

rth

11.3k7 gold badges58 silver badges88 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Amit Gupta Over a year ago

Thanks! hstack did the trick. I have updated the question with the solution.

Collectives™ on Stack Overflow

How to classify text pairs using scikit-learn?

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related