Truth-Conditional Captions for Time Series Data. EMNLP 2021

Code and Data for our EMNLP 2021 paper titled 'Truth-Conditional Captions for Time Series Data'.

Data

• processed_data/
• Contains
  • Synthetic time series data with NL annotations (pilot13final*)
  • Stock time series data with NL annotations (pilot16b*)

Code

• allennlp_series/
• Experiment scripts: TODO
• Pre-trained model files: TODO

Requirements

• python=3.6
• pip install -r requirements.txt file
• Additional dependency on cocoevals
  • Download cocoevals code from following URL: https://drive.google.com/file/d/17AO2x_8_ltHI9WoVbcZIvGNZZ7wzKvFo/view?usp=sharing
  • Move the folders to allennlp_series/training/metrics/

Citation

@inproceedings{jhamtani2021truth,
  title={Truth-Conditional Captioning of Time Series Data},
  author={Jhamtani, Harsh and Berg-Kirkpatrick, Taylor},
  booktitle={EMNLP},
  year={2021}
}