It is Keras implementation based on Kim's Convolutional Neural Networks for Sentence Classification paper .

Requirements

• Python 3
• Tensorflow > 1.6.0
• keras 2.1.6
• Numpy
• Flask (Optional,for server)

Training

Training procedure is described in training.ipynb file.

Print parameters:

 arguments in multichannel_drop:
        
  EMBEDDING_DIM
                        Dimensionality of character embedding (default: 100)
  FILTER_SIZES
                        hard-coded,Comma-separated filter sizes (default: '2,3,5')
  NUM_FILTERS
                        hard-coded,Number of filters per filter size (default: 100)
  
                        
   DROP
                        Dropout keep probability (default: 0.5),
  
  

Haven't used regularisation.

Pretrained Weights

Located in weights directory,along with embedding matrix used for data in data/ directory

Basic visualisation

Basic visualisation procedure is described in visualisation.ipynb file.

For interactive visualisation run server

Server

python server.py

Go to the address "127.0.0.1:8000/static/nlp.html" using browser .

References

• Convolutional Neural Networks for Sentence Classification