Residual Connected Sentence Encoder

This is a repo for Residual-connected sentence encoder for NLI. https://arxiv.org/abs/1708.02312
If you use this code as part of published research, please cite the following paper.

@article{nie2017shortcut,
  title={Shortcut-stacked sentence encoders for multi-domain inference},
  author={Nie, Yixin and Bansal, Mohit},
  journal={arXiv preprint arXiv:1708.02312},
  year={2017}
}

Try to follow the instruction below to successfully run the experiment.

1.Download the additional data.zip file, unzip it and place it at the root directory of this repo. Link for download data.zip file: DropBox Link

2.This repo is based on an old version of torchtext, the latest version of torchtext is not backward-compatible. We provide a link to download the old torchtext that should be used for this repo. Link: old_torchtext

3.Install the requirement:

python 3.6

torchtext # The one you just download. Or you can use the latest torchtext by fixing the SNLI path problem.
pytorch == 0.2.0
fire
tqdm
numpy
spacy

For the installation of torchtext, you can run the following script in the downloaded torchtext_backward_compatible directory (in step 2) using the python interpreter of your environment:

python setup.py install

To fully install spacy, you will need to run the following script.

pip install -U spacy
python -m spacy download en

Optionally, you can try to match the pip freeze file below to set up the same experiment environment.

certifi==2017.11.5
chardet==3.0.4
cymem==1.31.2
cytoolz==0.8.2
dill==0.2.7.1
en-core-web-sm==2.0.0
fire==0.1.2
ftfy==4.4.3
html5lib==1.0.1
idna==2.6
msgpack-numpy==0.4.1
msgpack-python==0.5.1
murmurhash==0.28.0
numpy==1.14.0
pathlib==1.0.1
plac==0.9.6
preshed==1.0.0
PyYAML==3.12
regex==2017.4.5
requests==2.18.4
six==1.11.0
spacy==2.0.5
termcolor==1.1.0
thinc==6.10.2
toolz==0.9.0
torch==0.2.0.post3
torchtext==0.1.1
tqdm==4.19.5
ujson==1.35
urllib3==1.22
wcwidth==0.1.7
webencodings==0.5.1
wrapt==1.10.11

4.There is a directory called saved_model At the root directory of this repo: This directory will be used for saving the models that produce best dev result.

Before running the experiments, make sure that the structure of this repo should be something like below.

.
├── config.py
├── data
│   ├── multinli_0.9
│   │   ├── multinli_0.9_dev_matched.jsonl
│   │   ├── multinli_0.9_dev_mismatched.jsonl
│   │   ├── multinli_0.9_test_matched_unlabeled.jsonl
│   │   ├── multinli_0.9_test_mismatched_unlabeled.jsonl
│   │   └── multinli_0.9_train.jsonl
│   ├── saved_embd.pt
│   └── snli_1.0
│       ├── README.txt
│       ├── snli_1.0_dev.jsonl
│       ├── snli_1.0_dev.txt
│       ├── snli_1.0_test.jsonl
│       ├── snli_1.0_test.txt
│       ├── snli_1.0_train.jsonl
│       └── snli_1.0_train.txt
├── model
│   └── res_encoder.py
├── saved_model
│   └── trained_model_will_be_saved_in_here.txt
├── setup.sh
├── torch_util.py
└── util
    ├── data_loader.py
    ├── dataset_util.py
    ├── __init__.py
    ├── mnli.py
    └── save_tool.py

5.Start training by run the script in the root directory.

source setup.sh
python model/res_encoder.py train_snli

6.After training completed, there will be a folder created by the script in the saved_model directory. The parameters of the model will be saved in that folder. The path of the model will be something like:

$DIR_TMP/saved_model/(TIME_STAMP)_[600,600,600]-3stack-bilstm-maxout-residual/saved_params/(YOUR_MODEL_WITH_DEV_RESULT)

Remember to change the bracketed part to the actual file name on your computer.

7.Now, you can evaluate the model on dev set again by running the script below.

python model/res_encoder.py eval (PATH_OF_YOUR_MODEL) dev # for evaluation on dev set
python model/res_encoder.py eval (PATH_OF_YOUR_MODEL) test # for evaluation on test set

Pretrained Model:
We also provide a link to download the pretrained model.
After downloading the pretrained model, you can run the script in step 7 for evaluation, however you need to keep the default parameter for pytorch to load the pretrained model.