BIMPM-pytorch

Re-implementation of BIMPM(Bilateral Multi-Perspective Matching for Natural Language Sentences, Zhiguo Wang et al., IJCAI 2017) on Pytorch

Results

Dataset: SNLI

Model	ACC(%)
Re-implementation	86.5
Baseline from the paper (Single BiMPM)	86.9

Dataset: Quora (provided by the authors)

Model	ACC(%)
Re-implementation	87.3
Baseline from the paper (Single BiMPM)	88.17

Note: I could not observe much gain from character embeddings. The implementation for the character embedding might be too naive to improve the performance.

Development Environment

OS: Ubuntu 16.04 LTS (64bit)
Language: Python 3.6.2.
Pytorch: 0.3.0

Requirements

Please install the following library requirements specified in the requirements.txt first.

nltk==3.2.4
torchtext==0.2.0
torch==0.3.0
tensorboardX==0.8

Training

python train.py --help

usage: train.py [-h] [--batch-size BATCH_SIZE] [--char-dim CHAR_DIM]
            [--char-hidden-size CHAR_HIDDEN_SIZE] [--data-type DATA_TYPE]
            [--dropout DROPOUT] [--epoch EPOCH] [--gpu GPU]
            [--hidden-size HIDDEN_SIZE] [--learning-rate LEARNING_RATE]
            [--max-sent-len MAX_SENT_LEN]
            [--num-perspective NUM_PERSPECTIVE] [--print-freq PRINT_FREQ]
            [--use-char-emb] [--word-dim WORD_DIM]

optional arguments:
  -h, --help                    show this help message and exit
  --batch-size BATCH_SIZE
  --char-dim CHAR_DIM
  --char-hidden-size CHAR_HIDDEN_SIZE
  --data-type DATA_TYPE         available: SNLI or Quora
  --dropout DROPOUT
  --epoch EPOCH
  --gpu GPU
  --hidden-size HIDDEN_SIZE
  --learning-rate LEARNING_RATE
  --max-sent-len MAX_SENT_LEN   max length of input sentences model can accept, if -1,
                                it accepts any length
  --num-perspective NUM_PERSPECTIVE
  --print-freq PRINT_FREQ
  --use-char-emb
  --word-dim WORD_DIM

Test

python test.py --help

usage: test.py [-h] [--batch-size BATCH_SIZE] [--char-dim CHAR_DIM]
           [--char-hidden-size CHAR_HIDDEN_SIZE] [--dropout DROPOUT]
           [--data-type DATA_TYPE] [--epoch EPOCH] [--gpu GPU]
           [--hidden-size HIDDEN_SIZE] [--learning-rate LEARNING_RATE]
           [--num-perspective NUM_PERSPECTIVE] [--use-char-emb]
           [--word-dim WORD_DIM] --model-path MODEL_PATH

optional arguments:
  -h, --help                    show this help message and exit
  --batch-size BATCH_SIZE
  --char-dim CHAR_DIM
  --char-hidden-size CHAR_HIDDEN_SIZE
  --dropout DROPOUT
  --data-type DATA_TYPE         available: SNLI or Quora
  --epoch EPOCH
  --gpu GPU
  --hidden-size HIDDEN_SIZE
  --learning-rate LEARNING_RATE
  --num-perspective NUM_PERSPECTIVE
  --use-char-emb
  --word-dim WORD_DIM
  --model-path MODEL_PATH

Note: You should execute test.py with the same hyperparameters, which are used for training the model you want to run.

galsang / BIMPM-pytorch