Convolutional Neural Networks for Sentence Classification

Pytorch re-implementation of Convolutional Neural Networks for Sentence Classification.

This is an unofficial implementation. There is the implementation by the authors, which is implemented on Theano.

Results

Baseline from the paper

Model	MR	SST-1	SST-2	Subj	TREC
random	76.1	45.0	82.7	89.6	91.2
static	81.0	45.5	86.8	93.0	92.8
non-static	81.5	48.0	87.2	93.4	93.6
multi-channel	81.1	47.4	88.1	93.2	92.2

Re-implementation

Model	MR	SST-1	SST-2	Subj	TREC
random	-	36.1	74.5	-	87.2
static	-	47.8	85.2	-	92.8
non-static	81.0	47.8	85.3	92.8	93.6
multi-channel	81.0	48.1	85.2	92.2	93.8

Development Environment

• OS: Ubuntu 16.04 LTS (64bit)
• Language: Python 3.6.6
• Pytorch: 0.4.0

Requirements

Please install the following library requirements first.

nltk==3.3
tensorboardX==1.2
torch==0.4.0
torchtext==0.2.3

Training

python train.py --help

usage: train.py [-h] [--batch-size BATCH_SIZE] [--dataset DATASET]
                [--dropout DROPOUT] [--epoch EPOCH] [--gpu GPU]
                [--learning-rate LEARNING_RATE] [--word-dim WORD_DIM]
                [--norm-limit NORM_LIMIT] [--mode MODE]
                [--num-feature-maps NUM_FEATURE_MAPS]

optional arguments:
  -h, --help            show this help message and exit
  --batch-size BATCH_SIZE
  --dataset DATASET     available datasets: MR, TREC, SST-1, SST-2, SUBJ
  --dropout DROPOUT
  --epoch EPOCH
  --gpu GPU
  --learning-rate LEARNING_RATE
  --word-dim WORD_DIM
  --norm-limit NORM_LIMIT
  --mode MODE           available models: rand, static, non-static,
                        multichannel
  --num-feature-maps NUM_FEATURE_MAPS

Note: