scikit-learn: how to use two different datasets for training and testing

I am trying to use different datasets as training set and test set respectively. But with the following code, I get:

File "main.py", line 84, in main_test X2 = tf_transformer.transform(word_counts2) File "/Library/Python/2.7/site-packages/sklearn/feature_extraction/text.py", line 1020, in transform n_features, expected_n_features)) ValueError: Input has n_features=1293 while the model has been trained with n_features=1625

def main_test(path = None):
    dir_path = path or 'dataset'
    files = sklearn.datasets.load_files(dir_path)
    util.refine_all_emails(files.data)
    word_counts = util.bagOfWords(files.data)
    tf_transformer = sklearn.feature_extraction.text.TfidfTransformer(use_idf=True)
    tf_transformer.fit(word_counts)
    X = tf_transformer.transform(word_counts)

    dir_path = 'testset'
    files2 = sklearn.datasets.load_files(dir_path)
    util.refine_all_emails(files2.data)
    word_counts2 = util.bagOfWords(files2.data)
    # tf_transformer = sklearn.feature_extraction.text.TfidfTransformer(use_idf=True)
    # tf_transformer.fit(word_counts2)
    X2 = tf_transformer.transform(word_counts2)

    clf = sklearn.svm.LinearSVC()

    test_classifier(X, files.target, clf, X2, files2.target, test_size=0.2, y_names=files.target_names, confusion=False)


def test_classifier(X, y, clf, X2, y2, test_size=0.4, y_names=None, confusion=False):
    X_train, X_test, y_train, y_test = X, X2, y, y2
    clf.fit(X_train, y_train)
    # clf.fit(X_test, y_test)
    y_predicted = clf.predict(X_test)

    print colored('Classification report:', 'magenta', attrs=['bold'])
    print sklearn.metrics.classification_report(y_test, y_predicted, target_names=y_names)