Implementing CNN Text Classification in MXNet

Recently, I have been learning mxnet for Natural Language Processing (NLP). I followed this official code in MXNet github. However, I find the official codes are too simple to run a whole process, so I changed it.

RNN text classification in MXNet is here.

The main difference with the official version

• Inference code were added, one can use his trained model to do prediction
• The MXNet version is 0.12.1, so some original functions may be deprecated
• Binary classification tasks were changed to multi-category tasks
• The codes about pretrained embedding were removed, data format were changed
• Label shape were changed to (batch_size,)

Data

training and validation data

two txt file, the format of each line is: <label> sentence.

• <pos> This is the best movie about troubled teens since 1998's whatever.
• <neg> This 10th film in the series looks and feels tired.

config data

one label a line, the number of labels is equals to total classes.

• pos
• neg

inference data

one sentence a line, without <label>

inference data with evaluation

the format of each line is: <label> sentence, like validation file

The data is recommended to be tokenized or segmented(Chinese).

Quick start

python cnn_model.py --train path/to/train.data --validate /path/to/validate.data --config /path/to/config

python inference.py --test python/to/inference.data --config /path/to/config --checkpoint 1

python inference.py --test python/to/inference-evaluation.data --config /path/to/config --checkpoint 1 --evaluation

References

• Implementing CNN + Highway Network for Chinese Text Classification in MXNet
• "Implementing a CNN for Text Classification in Tensorflow" blog post.
• Convolutional Neural Networks for Sentence Classification