Getting error while using gensim model in python

Question

I have made doc2vec file by training data using gensim model now while processing it. I am getting an error. I am running the below code:-

model = Doc2Vec.load('sentiment140.d2v')

if len(sys.argv) < 4:
    print ("Please input train_pos_count, train_neg_count and classifier!")
    sys.exit()

train_pos_count = int(sys.argv[1])
train_neg_count = int(sys.argv[2])
test_pos_count = 144
test_neg_count = 144

print (train_pos_count)
print (train_neg_count)

vec_dim = 100

print ("Build training data set...")
train_arrays = numpy.zeros((train_pos_count + train_neg_count, vec_dim))
train_labels = numpy.zeros(train_pos_count + train_neg_count)

for i in range(train_pos_count):
    prefix_train_pos = 'TRAIN_POS_' + str(i)
    train_arrays[i] = model.docvecs[prefix_train_pos]
    train_labels[i] = 1

for i in range(train_neg_count):
    prefix_train_neg = 'TRAIN_NEG_' + str(i)
    train_arrays[train_pos_count + i] = model.docvecs[prefix_train_neg]
    train_labels[train_pos_count + i] = 0


print ("Build testing data set...")
test_arrays = numpy.zeros((test_pos_count + test_neg_count, vec_dim))
test_labels = numpy.zeros(test_pos_count + test_neg_count)

for i in range(test_pos_count):
    prefix_test_pos = 'TEST_POS_' + str(i)
    test_arrays[i] = model.docvecs[prefix_test_pos]
    test_labels[i] = 1

for i in range(test_neg_count):
    prefix_test_neg = 'TEST_NEG_' + str(i)
    test_arrays[test_pos_count + i] = model.docvecs[prefix_test_neg]
    test_labels[test_pos_count + i] = 0


print ("Begin classification...")
classifier = None
if sys.argv[3] == '-lr':
    print ("Logistic Regressions is used...")
    classifier = LogisticRegression()
elif sys.argv[3] == '-svm':
    print ("Support Vector Machine is used...")
    classifier = SVC()
elif sys.argv[3] == '-knn':
    print ("K-Nearest Neighbors is used...")
    classifier = KNeighborsClassifier(n_neighbors=10)
elif sys.argv[3] == '-rf':
    print ("Random Forest is used...")
    classifier = RandomForestClassifier()

classifier.fit(train_arrays, train_labels)

print ("Accuracy:", classifier.score(test_arrays, test_labels))

I am getting a Keyerror - "TEST_POS_72"

I want to know what I am doing wrong.

Answer 1

The error means quite literally that no doc-vector with the key ('tag') TEST_POS_72 is part of the model. There mustn't have been any documents with that tag presented during training.

You can see a list of all known doc-tags in the model in model.docvecs.offset2doctag . If TEST_POS_72 isn't there, you can't access a doc-vector via model.docvecs['TEST_POS_72'] . (If that list is empty, then the doc-vectors were trained to be accessed by plain int keys – and model.docvecs[72] would be a more appropriate way to access a doc-vector.)

(Separately, Doc2Vec won't work well with tiny corpuses of a few hundred documents, and the warning in your screenshot "Slow version of gensim.models.doc2vec is being used" means that gensim's optimized C-compiled routines weren't part of the installation, and training will be 100x or more slower.)