Basketball and Football Win Probabilities

该数据集是一个包含4200场篮球和美式足球比赛胜率基准的集合，其中包含了1.7百万个时间步骤，每个步骤都有实值数据和对应该数据的自然语言事件描述。此外，该数据集还包括了NBA篮球和NFL美式足球中队伍A的胜率数据。为了消除记忆效应，每个模型在200个问题上的评估中，将实际的队伍和球员名称替换为通用标签。该数据集的规模达到了4200场比赛和1.7百万个时间步骤，其任务是从时间序列数据中推断自然语言事件。

This dataset is a collection of win rate benchmarks for 4200 basketball and American football matches, containing 1.7 million time steps. Each time step features real-valued data and corresponding natural language event descriptions. Additionally, the dataset includes win rate data for Team A in NBA basketball and NFL American football matches. To eliminate memory effects, actual team and player names are replaced with generic labels during the evaluation of each model across 200 questions. The dataset encompasses 4200 matches and 1.7 million time steps, with the core task being the inference of natural language events from time-series data.