luxury-lakehouse/football2vec-360-training-data

Football2Vec 360训练数据集是一个用于训练Football2Vec 360-Enriched模型的数据集，包含来自StatsBomb 360开放数据的323场职业足球比赛的约200万次动作。每个数据行代表一个球员在一场比赛中的动作序列，包括SPADL标记化的事件和StatsBomb 360的冻结帧上下文。冻结帧上下文是该数据集的关键特点，它使得Deep Sets编码器能够学习球员相对于周围球员的行为，捕捉到仅凭动作序列无法揭示的空间决策信息。数据集还包括球员标识符、比赛标识符、比赛和赛季信息、球员位置组等字段。该数据集适用于Transformer + Deep Sets训练、空间上下文研究、自定义嵌入模型训练和消融研究等用途。

The Football2Vec 360 Training Dataset is a curated dataset designed for training the Football2Vec 360-Enriched model. It contains approximately 2 million action instances across 323 professional soccer matches sourced from the StatsBomb 360 Open Data. Each data row represents an action sequence of a single player during a match, including SPADL-tokenized events and StatsBomb 360 freeze-frame context. Freeze-frame context is the core distinctive feature of this dataset: it enables the Deep Sets encoder to learn player behaviors relative to surrounding teammates and opponents, capturing spatial decision-making information that cannot be revealed solely by action sequences. The dataset also includes fields such as player identifiers, match identifiers, match and season metadata, and player position groups. This dataset is suitable for use cases including Transformer + Deep Sets training, spatial context research, custom embedding model training, and ablation studies.