table-annotation-using-deep-learning

Team Project HWS2022 Table Annotation using deep learning

Reproduction

To reproduce the experiments:

• Build the conda environment

  The conda environment tp-dws.yml is available at the root directory. Install using conda env create -f tp-dws.yml

• Download the data for SOTAB

  Remain at root directory and execute download.sh

• Preprocess data

  Redirect to respective folders for Column Type Annotation (CTA) and Column Property Annotation (CPA) under experiments_final_phase/. Run the create_new_dataset python script to preprocess the respective data python create_new_dataset.py

• Run experiments

  Example reproduction code is available at run.py

To reproduce the TURL experiments:

• Download the Wikitables data

  Download rom https://github.com/sunlab-osu/TURL. Redirect to the respective directory of experiments_turl/cta or experiments_turl/cpa and execute turl_create_cta_pickle.ipynb or turl_create_cpa_pickle.ipynb

• Run experiments

  The workflow is similar to our workflow for SOTAB benchmark

For additional experiments:

• Subtables model The code is available in experiments_final_phase/cpa/create_subtables.py to create the subtables fpr the CPA task