USPTO-15K

.

该数据集是美国专利商标局（USPTO）数据集的一个子集，包含了15,000个反应，其中包括了1,700个最常见模板所涵盖的反应。此外，该数据集用于与基于模板的高性能方法比较模型准确性。其规模为15,000个反应，分为10,500个用于训练，1,500个用于开发，以及3,000个用于测试。所承担的任务是对反应预测模型的比较评估。

This dataset is a subset of the United States Patent and Trademark Office (USPTO) dataset, comprising 15,000 chemical reactions, including those covered by the 1,700 most prevalent reaction templates. Furthermore, this dataset is designed for comparative assessment of model accuracy against template-based high-performance methods. In total, it contains 15,000 reactions, which are split into 10,500 for training, 1,500 for development, and 3,000 for testing. The core task of this dataset is the comparative evaluation of reaction prediction models.