zhykoties/time-series-language-alignment

TS-Insights数据集是论文《Insight Miner: A Time Series Analysis Dataset for Cross-Domain Alignment with Natural Language》的官方数据集，由Google X的Project Mineral于2023年完成。这是第一个大规模通用领域数据集，旨在将时间序列数据与自然语言描述对齐。该数据集支持训练大型多模态模型（LMMs），以理解时间序列作为一种新的输入模态，并生成全面的文本洞察，特别关注趋势分析、季节性和残差。数据集包含来自Monash时间序列存档的20个不同数据集，涵盖多个领域和粒度。数据集通过趋势生成和增强技术进行扩展，以提高模型的鲁棒性。数据集结构分为训练和评估存档，详细说明了每个存档的内容和组织方式。

TS-Insights is the official dataset for the paper "Insight Miner: A Time Series Analysis Dataset for Cross-Domain Alignment with Natural Language". This work is done by Project Mineral from Google X in 2023. It is the first large-scale general-domain dataset designed to align time-series data with natural language descriptions. The dataset supports the training of Large Multimodal Models (LMMs) to understand time series as a new input modality and generate comprehensive textual insights, specifically focusing on trend analysis, seasonality, and residuals. The dataset is constructed from 20 distinct datasets from the Monash Time Series Archive, covering diverse domains and granularities. It includes trend generation and augmentation techniques to scale the dataset and improve model robustness. The dataset is organized into training and evaluation archives, with specific details on the contents and structure of each.