Source code for " Number-Enhanced Representation with Hierarchical Recursive Tree Decoding for Math Word Problem Solving".
Please note that the currently available code has not undergone a thorough refinement process. We intend to upload a more polished and robust version of the code after the paper's publication.
There are four datasets used in this work, Math23K, MAWPS, SVAMP and MathQA. Preprocessed data can be directly accessed in the ./data but need to decompression.
All the training logs for the experiments in the paper and the corresponding model checkpoints can be accesses in this link.
python == 3.8.xLTP == 4.1.5.post2for the Chinese datasettransformers == 4.7.xCoreNLP == 4.4.0for the English dataset
Do the following: All the commands are listed in the scripts directory. Before rerunning the training process, please make sure the pretrained data are downloaded and untared into datasets directory; before doing the test, please make sure the model checkpoints are downloaded and untared into cache models directory.
Followers can rerun the training process by the run.py scripts with train in their name, or reproduce the experiment results by scripts with test.py in their name.