AnDuShu

An inclusive natural language to code library built over allennlp 2.

---

This repository is using code from the following resources:

• allennlp-models
• allennlp_sempar
• allennlp-semparse

In this repo, we adapt the code to the latest allennlp version.

---

ATIS

PYTHONPATH=src ANNOTATION_DIR=$PWD/data/annotations/atis/ \
  allennlp train configs/atis/seq2seq/defaults.jsonnet \
  -s data/output/atis/seq2seq --include-package andushu

PYTHONPATH=src ANNOTATION_DIR=$PWD/data/annotations/atis/ \
  allennlp predict \
  --output-file=data/output/atis/seq2seq/seq2seq.jsonl \
  --predictor seq2seq \
  --include-package andushu \
  data/output/atis/seq2seq/model.tar.gz \
  data/annotations/atis/atis_test.jsonl 

GEOQUERY

PYTHONPATH=src ANNOTATION_DIR=$PWD/data/annotations/geoquery/ \
  allennlp train configs/geoquery/seq2seq/defaults.jsonnet \
  -s data/output/geoquery/seq2seq --include-package andushu

PYTHONPATH=src ANNOTATION_DIR=$PWD/data/annotations/geoquery/ \
  allennlp predict \
  --output-file=data/output/geoquery/seq2seq/seq2seq.jsonl \
  --predictor seq2seq \
  --include-package andushu data/output/geoquery/seq2seq/model.tar.gz \
  data/annotations/geoquery/geo_test.jsonl 

Math Word Problems

This repo includes the code for

@article{Tan2021InvestigatingMW,
  title={Investigating Math Word Problems using Pretrained Multilingual Language Models},
  author={Minghuan Tan and Lei Wang and Lingxiao Jiang and Jing Jiang},
  journal={ArXiv},
  year={2021},
  volume={abs/2105.08928}
}

Annotations

Annotations used for this paper can be found at

• 百度网盘

• GoogleDrive.

Training

Training Math23K using bert-base-multilingual-cased.

CUDA_VISIBLE_DEVICES=2 PROJECT=math2tree SUB_PROJECT=math23k CONFIG=transformer_vocab SPACY_LANGUAGE=zh MODEL_NAME=bert-base-multilingual-cased OP_TYPE=disallow_pow bash docker_train.sh

Training over MathQA-Adapted without Pow.

CUDA_VISIBLE_DEVICES=2 PROJECT=math2tree SUB_PROJECT=mathqa CONFIG=transformer_vocab SPACY_LANGUAGE=zh MODEL_NAME=bert-base-multilingual-cased OP_TYPE=disallow_pow bash docker_train.sh

Training over MathQA-Adapted with Pow.

CUDA_VISIBLE_DEVICES=2 PROJECT=math2tree SUB_PROJECT=mathqa CONFIG=transformer_vocab SPACY_LANGUAGE=zh MODEL_NAME=bert-base-multilingual-cased OP_TYPE=allow_pow bash docker_train.sh

Training over MathXLing without Pow.

CUDA_VISIBLE_DEVICES=2 PROJECT=math2tree SUB_PROJECT=mathxling CONFIG=transformer_vocab SPACY_LANGUAGE=zh MODEL_NAME=bert-base-multilingual-cased OP_TYPE=disallow_pow bash docker_train.sh

Training over MathXLing with Pow.

CUDA_VISIBLE_DEVICES=2 PROJECT=math2tree SUB_PROJECT=mathxling CONFIG=transformer_vocab SPACY_LANGUAGE=zh MODEL_NAME=bert-base-multilingual-cased OP_TYPE=allow_pow bash docker_train.sh

Evaluation

CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/mathqa/transformer_vocab_en_disallow_pow_bert-base-multilingual-cased evaluate
CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/mathqa/transformer_vocab_en_allow_pow_bert-base-multilingual-cased evaluate

CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/mathqa/transformer_vocab_en_disallow_pow_bert-base-multilingual-cased evaluate 
CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/math23k/transformer_vocab_zh_disallow_pow_bert-base-multilingual-cased evaluate
 
CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/math23k/transformer_vocab_zh_allow_pow_bert-base-multilingual-cased evaluate
CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/math23k/transformer_vocab_en_allow_pow_bert-base-multilingual-cased evaluate

CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/math23k/transformer_vocab_en_disallow_pow_bert-base-multilingual-cased evaluate

CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/mathxling/transformer_vocab_zh_sallow_pow_bert-base-multilingual-cased evaluate
CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/mathxling/transformer_vocab_zh_allow_pow_bert-base-multilingual-cased evaluate

CUDA_VISIBLE_DEVICES=7 bash docker_eval.sh math2tree data/output/mathxling/transformer_vocab_zh_allow_pow_xlm-roberta-base evaluate