logo-lab/trl-ctbench
收藏资源简介:
TRL-CTbench是一个用于表格数据表示学习的基准测试套件,包含27个配置,覆盖多种任务如检索式联合搜索、模式匹配、列类型预测、表问答、关系提取等。数据集规模在1M到10M之间,使用CC-BY-SA-4.0许可证。它提供了多个子数据集,如ckan_subset、ecb_union、nq_tables、sato、valentine等,每个配置针对不同的表格数据处理任务,支持训练、验证和测试分割。数据集适用于评估表格数据的表示学习模型,并包含快速使用示例。
TRL-CTbench is a benchmark suite for tabular data representation learning, which includes 27 configurations covering various tasks such as retrieval-based joint search, pattern matching, column type prediction, table question answering, relation extraction and more. The dataset has a scale ranging from 1M to 10M, and is released under the CC-BY-SA-4.0 license. It provides multiple sub-datasets like ckan_subset, ecb_union, nq_tables, sato, valentine and others. Each configuration targets different tabular data processing tasks, and supports training, validation and test splits. This dataset is suitable for evaluating tabular data representation learning models, and also contains quick usage examples.




