NHL Game Data

### Context

For some, statistical analysis increases the enjoyment of sport. My long-term aim is to build an automated database driven advanced stats website like many that have come and gone before however that first starts with statistical analysis of the game to assess what and how much individual actions contribute to the outcome of the game.

The data represents all the official metrics measured for each game in the NHL in the past 6 years. I intend to update it semi-regularly depending on development progress of my database server.

### Content

This is a mostly rational database, please refer to the "table_realtionships.jpg" for details on how the tables can be joined. This is not just the results and player stats of NHL games but also details on individual plays such as shots, goals and stoppages including date & time and x,y coordinates.

The dataset is incomplete, there are some games where no plays information is available on NHL.com. It is rare and I do not know the reasons.

### Acknowledgements

Thanks to Kevin Sidwar who began [documenting the still un-documented NHL stats API][1] which was used to gather this data.

### Inspiration

Compared to other sports, advanced statistics in Hockey are still in infancy. It has been suggested that the best models [can only predict the winner 62% of the time][2] due to variances in talent and "puck luck".

I would like to believe feature engineering and a suitably trained model can account for some of this variance and beat this seemingly low target.

Otherwise, what metrics can be developed to provide better indications than Corsi & Fenwick?

[1]: https://www.kevinsidwar.com/iot/2017/7/1/the-undocumented-nhl-stats-api
[2]: https://www.nhlnumbers.com/2013/08/01/machine-learning-and-hockey-is-there-a-theoretical-limit-on-predictions

{'Context': '对于部分人而言，统计学分析能够提升体育运动的观赏乐趣。吾之长远之志，在于构建一个自动化数据库驱动的先进统计数据网站，如同诸多前赴后继者所为之事。然而，此一目标之实现，首当其冲者，即为对比赛进行统计学分析，以评估个体动作对比赛结果所贡献之程度及多少。

此数据集涵盖过去六年中美国国家冰球联盟（NHL）每场比赛所测量的所有官方指标。吾意欲根据数据库服务器之发展进度，定期或不定期地更新之。
', 'Content': '此数据库主要为理性之数据库，具体表格之间的关系，请参阅“table_realtionships.jpg”以获取详细信息。此非仅NHL比赛之结果与球员统计数据，亦包括个人动作之详细信息，如射门、进球及停赛等，包括日期、时间以及x、y坐标。

数据集尚不完整，存在部分比赛在NHL.com上无可用之动作信息。此情况虽属罕见，然其成因吾亦未知。
', 'Acknowledgements': '感谢Kevin Sidwar，他开始[记录尚待完善的NHL统计数据API][1]，该API被用于收集此数据。
', 'Inspiration': '相较于其他运动，冰球之高级统计数据仍处于初级阶段。有观点认为，最佳模型[仅能在62%的时间内预测出胜者][2]，这是由于人才差异及“球运”之影响。

吾愿相信，通过特征工程与适当训练的模型，可以解释部分此类差异，并超越这一看似低企的目标。
否则，又能开发出哪些指标，以提供比Corsi与Fenwick更佳的指示呢？

[1]: https://www.kevinsidwar.com/iot/2017/7/1/the-undocumented-nhl-stats-api
[2]: https://www.nhlnumbers.com/2013/08/01/machine-learning-and-hockey-is-there-a-theoretical-limit-on-predictions
'}