vmando / cnn_sentence Goto Github PK

This is a fork from Yoon Kim repository for sentence classification using Convolutional Neural Networks.

The input to the network are sentences where each token has been labeled with a binary tag (for instance a Named Entity Recognition tag). The label of each token is then appended to its word vectors and used during the training of the CNN.

Code is written in Python (2.7) and requires Theano (0.7).

Using the pre-trained word2vec vectors will also require downloading the binary file from https://code.google.com/p/word2vec/

The raw data is assumed to be the (tab-separated) juxaposition of the sentence and its associated label sequence. See the files dataset.pos and dataset.neg for an example.

To process the raw data, run

python process_data.py

where path points to the word2vec binary file (i.e. GoogleNews-vectors-negative300.bin file). This will create a pickle object called mr.p in the same folder.

Example commands:

THEANO_FLAGS=mode=FAST_RUN,device=cpu,floatX=float32 python conv_net_sentence.py -nonstatic -rand
THEANO_FLAGS=mode=FAST_RUN,device=cpu,floatX=float32 python conv_net_sentence.py -static -word2vec
THEANO_FLAGS=mode=FAST_RUN,device=cpu,floatX=float32 python conv_net_sentence.py -nonstatic -word2vec

This will run the CNN-rand, CNN-static, and CNN-nonstatic models respectively in the paper.

GPU will result in a good 10x to 20x speed-up, so it is highly recommended. To use the GPU, simply change device=cpu to device=gpu (or whichever gpu you are using). For example:

THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32 python conv_net_sentence.py -nonstatic -word2vec