Skip to content

Latest commit

 

History

History
48 lines (31 loc) · 2.86 KB

README.md

File metadata and controls

48 lines (31 loc) · 2.86 KB
Raw

Run the code

This directory contains code to train and test the CMR model. We provide two experimental conditions:

  • Toy data: This one does NOT require generating the data using scripts ../data, as a toy dataset is already provided. This only requires executing the steps (1), (2) and (4) below. This lets the user quickly run the entire pipeline end-to-end to more easily understand how it works. This is of course only useful for understanding purposes, as the small size of the training data will of course cause the generated outputs to be of very poor quality.
  • Full data: This runs on the full data, which requires first running scripts ../data according to the instructions in that folder. This creates files in data/raw/full. After that raw data has been created, please run all the steps listed below.

1) Setup Environment

  1. python3.6
  2. install requirements:

    pip install -r requirements.txt

  3. You might need to download the en module for spacy

    python -m spacy download en # default English model (~50MB)
    python -m spacy download en_core_web_md # larger English model (~1GB)

Or pull our published docker: allenlao/pytorch-allennlp-rt
Hints:

If it is your first time to use docker, please refer the link for the usage:
https://docs.docker.com/get-started/

Regrading PyTorch versions, please refer:
https://pytorch.org/get-started/previous-versions/

2) Download files too large to fit as supplementary materials

Please download this tar gzip file and extract it under model. This should create a model/data folder. If you extracted in a different location, please make sure it is renamed as model/data.

3) Process data (Skip this step if you just want to train on toy data)

  • Converting the raw text file to id json file for training, run sh run_preprocessing.sh

  • The output_path is /data/processed/full. You can modify the output_path by changing --data_dir.

4) To train the model

  • Run run_san_data_weighted.sh on toy data by default.

  • If you want to train on full data, please modify --data_dir from /data/processed/toy to /data/processed/full .

5) To download the model checkpoint

Please download this checkpoint file and extract it under checkpoint. Then add --resume './checkpoint/san_checkpoint.pt‘ into sh run_preprocessing.sh

(Experimental) Interactive Model Demo

Given a query to the model, the demo extracts the most relevant entity from the query, gets corresponding grounding from Wikipedia (this step currently takes some time), and then generates a responses grounded on that query.

  • Run interactive_demo.py with relevant configuration flags to play with the model