/latent-treelstm

Cooperative Learning of Disjoint Syntax and Semantics

Primary LanguagePythonMIT LicenseMIT

Cooperative Learning of Disjoint Syntax and Semantics

This code was used to obtain the results described in the paper:

Cooperative Learning of Disjoint Syntax and Semantics
Serhii Havrylov, Germán Kruszewski, Armand Joulin
Presented at NAACL2019

ListOps

  1. Download ListOps dataset. URLs of the original dataset and an extrapolation test set can be found in data/listops/external/urls.txt file.
  2. Run python listops/data_preprocessing/split.py to split the dataset into the train, the valid and the test sets. Make sure that you have downloaded the dataset and the files are present in the data/listops/external folder.
  3. Build vocabulary using python listops/data_preprocessing/build_vocab.py.
  4. Run python listops/ppo/train_ppo_model.py or python listops/reinforce/train_reinforce_model.py to train the model with PPO or REINFORCE estimators.

SST

  1. Run python sst/ppo/train_ppo_model.py to train the model using SST-2 or SST-5 datasets.

NLI

  1. Download SNLI and MultiNLI datasets and extract corresponding archives to data/nli folder. URLs can be found in data/nli/external/urls.txt file.
  2. Run python nli/data_preprocessing/preprocess.py to preprocess dataset and generate vocabulary files.
  3. Run python nli/ppo/train_ppo_model.py to train the model using SNLI or MultiNLI datasets.

The code is tested with Python 3.6.3 and PyTorch 1.0.1.

License

Latent-TreeLSTM is MIT licensed. See the LICENSE file for details.