/retned

Work for R252 on A Retrieve-and-Edit Framework for Predicting Structured Outputs (NeurIPS2018)

Primary LanguagePython

Data processing:

  • for r252 dataset, call data_processing.py r252_build_dataset("r252")
  • for hstone dataset, call hstone_process.py hstone_build_dataset("hstone")

Training: python rnn_attn_reted.py

  • arguments --fourthofdata, --halfdata, --threefourthsofdata specify whether to run the model with 25%, 50% or 75% of the data
  • argument --resume_ret specifies whether a retriever model has already been saved and loads its weights
  • argument --dataset_name specifies which dataset to run the model with. Implemented 'r252', 'hstone'