Skip to content
Math word problems sovler with equation normalization published in EMNLP 2018
Python Jupyter Notebook Shell
Branch: master
Clone or download
SumbeeLei Update DecoderRNN_3.py
Uncomment attention module.
Latest commit 78156d6 Sep 24, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data first commit Nov 26, 2018
experiment update 1126 Nov 26, 2018
scripts first commit Nov 26, 2018
src Update DecoderRNN_3.py Sep 24, 2019
README.md Update README.md Jan 27, 2019
data_process.ipynb first commit Nov 26, 2018

README.md

Code for math word problems solver using Bi-LSTM with equation normalization


We rewrote the code, and at present we only finished two normalization methods in this code (test acc: 65.5%), cause I haven't found previous preprocessing code, I will complement another normalization method as soon as possible.(but we provide the original norm files, so you can use directly) Thus, I directly used the previous templates with EN to do the entire experiments (test acc: ~67%).

This code requires python 3.5, pytorch 0.4 and some common python tools.

Code for data processing is in the data_process.ipynb

For SNI, we directly use the results from the paper "Deep Neural Solver for Math Word Problems". You can find relevant data from ./data/sni_dict.json

sh ./script/exe_post.sh model_dir

The training accuracy, and test accuracy will be printed. (you can randomly sample 1000 problems as validation set to help you tune your hyperparameters, and then train the model based on all training problems with the hyperparameters you choose)

References


You can’t perform that action at this time.