MiMoTable Dataset

December 17, 2024 ยท View on GitHub

This is the dataset MiMoTable in paper "MiMoTable: A Multi-scale Spreadsheet Benchmark with Meta Operations for Table Reasoning".

Spreadsheet Files

The data/{language}/spreadsheet directory contains the spreadsheet files.

Questions and Answers

The data/{language}/problems.json file contains the questions, answers and other related infos. Each line contains one sample. The fields in each sample are described as below.

  • spreadsheet_list: the paths of spreadsheets used to answer the question.
  • table_type: the type of tables described in paper.
  • table_difficulty: the table difficulty, three level, simple/medium/hard
  • question
  • meta_operation_list: the meta operations of question.
  • question_difficulty: the question difficulty described in paper.
  • answer
  • output_files: the output files of Edit and Visualize, xlsx or png.

The src directory contains the code for building related prompt.

  • meta_operation_classification.py
  • model_inference.py
  • performance_evaluation.py