hugochan / rl-based-graph2seq-for-nqg Goto Github PK

Code & data accompanying the ICLR 2020 paper "Reinforcement Learning Based Graph-to-Sequence Model for Natural Question Generation"

License: Apache License 2.0

Python 100.00%

question-generation text-generation graph2seq reinforcement-learning deep-learning pytorch graph-neural-networks

rl-based-graph2seq-for-nqg's Introduction

RL-based-Graph2Seq-for-NQG

Code & data accompanying the ICLR 2020 paper "Reinforcement Learning Based Graph-to-Sequence Model for Natural Question Generation".

Get started

Prerequisites

This code is written in python 3. You will need to install a few python packages in order to run the code. We recommend you to use virtualenv to manage your python packages and environments. Please take the following steps to create a python virtual environment.

If you have not installed virtualenv, install it with pip install virtualenv.
Create a virtual environment with virtualenv venv.
Activate the virtual environment with source venv/bin/activate.
Install the package requirements with pip install -r requirements.txt.

Run the model

Download the preprocessed data from squad-split1 and squad-split2. And put the data under the root directory. So the file hierarchy will be like: data/squad-split1 and data/squad-split2.
Run the model
```
python main.py -config config/squad_split1/graph2seq_static_bert_finetune_word_70k_0.4_bs_60.yml
```
Note that you can specify the output path by modifying out_dir in a config file. If you want to finetune a pretrained model, you can specify the path to the pretrained model by modifying pretrained and you need to set out_dir to null. If you just want to load a pretrained model and evaluate it on a test set, you need to set both trainset and devset to null.

Finetune the model using RL

python main.py -config config/squad_split1/rl_graph2seq_static_bert_finetune_word_70k_0.4_bs_60.yml

Reference

If you found this code useful, please consider citing the following paper:

Yu Chen, Lingfei Wu and Mohammed J. Zaki. "Reinforcement Learning Based Graph-to-Sequence Model for Natural Question Generation." In Proceedings of the 8th International Conference on Learning Representations (ICLR 2020), Addis Ababa, Ethiopia, Apr. 26-30, 2020.

@inproceedings{chen2019reinforcement,
author    = {Chen, Yu and Wu, Lingfei and Zaki, Mohammed J.},
title     = {Reinforcement Learning Based Graph-to-Sequence Model for Natural Question Generation},
booktitle = {Proceedings of the 8th International Conference on Learning Representations},
month = {Apr. 26-30,},
year      = {2020}}

rl-based-graph2seq-for-nqg's People

Contributors

Stargazers

Watchers

Forkers

ammieqi lambert-hpx zakimjz passenger20 zeta1999 czh513 canyuchen wly-thu andrei-volkau marwaallam know-nothing8 luomuqinghan aniruddha-ju theonll virajpwr crystalsixone johane-takeuchi talha1503 linshaoxin-maker

rl-based-graph2seq-for-nqg's Issues

Question: which input fields are actually required for the model?

I am sending an example.

I trained a model. I would like to try it on my own data. Could you please clarify which fields are actually required for the model? It might be possible that I don't need to provide all of them while shaping my own data as per the format below.

{
  "text3": "Twins",
  "text1": "twins -lrb- 1988 -rrb- , a comedy with danny devito , also proved successful .",
  "text2": "What's the title of the comedy movie Schwarzenegger starred in with Danny DeVito in 1988?",
  "annotation3": {
    "raw_text": "Twins",
    "toks": "Twins",
    "POSs": "NNS",
    "positions": "0-0-5",
    "NERs": "O"
  },
  "id": "56de8c374396321400ee2a11",
  "annotation2": {
    "raw_text": "What's the title of the comedy movie Schwarzenegger starred in with Danny DeVito in 1988?",
    "toks": "What 's the title of the comedy movie Schwarzenegger starred in with Danny DeVito in 1988 ?",
    "POSs": "WP VBZ DT NN IN DT NN NN NNP VBD IN IN NNP NNP IN CD .",
    "positions": "0-0-4 1-4-6 2-7-10 3-11-16 4-17-19 5-20-23 6-24-30 7-31-36 8-37-51 9-52-59 10-60-62 11-63-67 12-68-73 13-74-80 14-81-83 15-84-88 16-88-89",
    "NERs": "O O O O O O O O PERSON O O O PERSON PERSON O DATE O"
  },
  "annotation1": {
    "raw_text": "twins -lrb- 1988 -rrb- , a comedy with danny devito , also proved successful .",
    "toks": "twins -lrb- 1988 -rrb- , a comedy with danny devito , also proved successful .",
    "POSs": "NNS JJ CD NN , DT NN IN JJ NN , RB VBD JJ .",
    "positions": "0-0-5 1-6-11 2-12-16 3-17-22 4-23-24 5-25-26 6-27-33 7-34-38 8-39-44 9-45-51 10-52-53 11-54-58 12-59-65 13-66-76 14-77-78",
    "NERs": "O O DATE O O O O O O O O O O O O",
    "graph": {
      "g_features": [
        "twins",
        "-lrb-",
        "1988",
        "-rrb-",
        ",",
        "a",
        "comedy",
        "with",
        "danny",
        "devito",
        ",",
        "also",
        "proved",
        "successful",
        "."
      ],
      "g_adj": {
        "0": [
          {
            "node": 1,
            "edge": "punct"
          },
          {
            "node": 2,
            "edge": "appos"
          },
          {
            "node": 3,
            "edge": "punct"
          },
          {
            "node": 4,
            "edge": "punct"
          },
          {
            "node": 6,
            "edge": "appos"
          },
          {
            "node": 10,
            "edge": "punct"
          }
        ],
        "6": [
          {
            "node": 5,
            "edge": "det"
          },
          {
            "node": 7,
            "edge": "prep"
          }
        ],
        "7": [
          {
            "node": 9,
            "edge": "pobj"
          }
        ],
        "9": [
          {
            "node": 8,
            "edge": "amod"
          }
        ],
        "12": [
          {
            "node": 0,
            "edge": "nsubj"
          },
          {
            "node": 11,
            "edge": "advmod"
          },
          {
            "node": 13,
            "edge": "oprd"
          },
          {
            "node": 14,
            "edge": "punct"
          }
        ]
      },
      "num_edges": 14
    }
  }
}

loss data

Hello, can you give me a copy of the output file corresponding to the running path after the main.py command？

`rl_wmd_ratio` not found

Hi, I tried to run the exact code in the readme, but encountered this error.

Read the paper, and it's a great work by the way. Thanks for sharing the code.

Traceback (most recent call last):
  File "main.py", line 57, in <module>
    main(config)
  File "main.py", line 22, in main
    model = ModelHandler(config)
  File "C:\Users\User\Desktop\G2S_NQG\RL-based-Graph2Seq-for-NQG\src\core\model_handler.py", line 73, in __init__
    self.model = Model(config, train_set)
  File "C:\Users\User\Desktop\G2S_NQG\RL-based-Graph2Seq-for-NQG\src\core\model.py", line 69, in __init__
    if config['rl_wmd_ratio'] > 0:
KeyError: 'rl_wmd_ratio'