全部回帖
为什么没有苦力
为什么没有苦力
詹姆斯来的第一年也没进
受伤了怎么说，而且那年没有附加赛规则有就完全进附加赛了
受伤了怎么说，而且那年没有附加赛规则有就完全进附加赛了
穆雷40%命中率，无话可说
比哈登强
比哈登强
笑疯了
笑疯了
大雷
请问跟库里有啥关系
勇士替补真的断档，看着也没啥明星球员
勇士替补真的断档，看着也没啥明星球员
OBPM 库里联盟第二
不黑庫里，但Adjusted Plus/Minus系的高階數據現在在球探與分析員中的專業參考度並不高，而兩代bpm及其衍生數據類型除了用於跨時空縱向比對球員基本不會出現在有價值的分析報告里。
Why might APM be giving such non-intuitive results in some cases? Isit really discovering that the true value of various players is dramatically different from their commonly perceived value? In some cases, it may be doingso, but there are reasons to be concerned about APM’s accuracy. Withinthe basketball analytics community, much attention has been focused onthe issue of multicollinearity, which in this context corresponds to situationswhere pairs of players are very frequently or very rarely on the floor at thesame time. However, it is the opinion of the author that multicollinearityis only part of the story behind the struggles of the APM technique. Thereis a more general phenomenon at play, known amongst the statistics andmachine learning community as overfitting [1].Overfitting occurs when a model fits the training data too precisely, i.e.,in such a way that the fluky pecularities and noise of the data are fit, sothat the model’s predictive performance on future data is degraded. As asimple example, imagine a situation where a rookie NBA player has a FG%of 67% over the first 5 games of the season. Suppose the task is to predictthe player’s FG% over the course of the remaining 77 games. It should beintuitively obvious that the estimate which best fits the available data, 67%,is unlikely to be the optimal prediction for the entire season. The naive use2of such an estimate for the purposes of prediction would be an instance ofoverfitting. A smarter approach would be to combine the data frfrom the 5games with prior information about the typical distribution of FG% whichNBA players achieve over the course of the season.While it is true that APMmodels are often estimated over an entire season of data (or even multipleseasons) it is nonetheless the case that standard linear regression is proneto overfitting, since the data is so noisy and since so many parameters (onefor nearly every player in the league) need to be simultaneously estimated.
不黑庫里，但Adjusted Plus/Minus系的高階數據現在在球探與分析員中的專業參考度並不高，而兩代bpm及其衍生數據類型除了用於跨時空縱向比對球員基本不會出現在有價值的分析報告里。
Why might APM be giving such non-intuitive results in some cases? Isit really discovering that the true value of various players is dramatically different from their commonly perceived value? In some cases, it may be doingso, but there are reasons to be concerned about APM’s accuracy. Withinthe basketball analytics community, much attention has been focused onthe issue of multicollinearity, which in this context corresponds to situationswhere pairs of players are very frequently or very rarely on the floor at thesame time. However, it is the opinion of the author that multicollinearityis only part of the story behind the struggles of the APM technique. Thereis a more general phenomenon at play, known amongst the statistics andmachine learning community as overfitting [1].Overfitting occurs when a model fits the training data too precisely, i.e.,in such a way that the fluky pecularities and noise of the data are fit, sothat the model’s predictive performance on future data is degraded. As asimple example, imagine a situation where a rookie NBA player has a FG%of 67% over the first 5 games of the season. Suppose the task is to predictthe player’s FG% over the course of the remaining 77 games. It should beintuitively obvious that the estimate which best fits the available data, 67%,is unlikely to be the optimal prediction for the entire season. The naive use2of such an estimate for the purposes of prediction would be an instance ofoverfitting. A smarter approach would be to combine the data frfrom the 5games with prior information about the typical distribution of FG% whichNBA players achieve over the course of the season.While it is true that APMmodels are often estimated over an entire season of data (or even multipleseasons) it is nonetheless the case that standard linear regression is proneto overfitting, since the data is so noisy and since so many parameters (onefor nearly every player in the league) need to be simultaneously estimated.
原来还有一项能进前四啊 不错
那咋了西部第二有问题吗
那咋了西部第二有问题吗
为什么要除了，勇士这三年也就24没进季后赛，好歹进附加赛了，不像某队4个75大赛季都没结束就out了
威什么呢？人家詹眉两人就冠军了，你来了怎么反而拖后腿了，这是威什么呢？瓜哥最后一舞更是被毁了，四个75大谁打的最差呢？
威什么呢？人家詹眉两人就冠军了，你来了怎么反而拖后腿了，这是威什么呢？瓜哥最后一舞更是被毁了，四个75大谁打的最差呢？
图图也是好起来了，居然只排在约老师后面了。
图图也是好起来了，居然只排在约老师后面了。
OBPM 库里联盟第二
同是一代高階數據比較bpm系，以TS與per來評析更明了庫里的比賽影響力與技術特點
同是一代高階數據比較bpm系，以TS與per來評析更明了庫里的比賽影響力與技術特點
原来还有一项能进前四啊 不错
Improved NBA Adjusted +/- Using Regularizationand Out-of-Sample Testing
Joseph Sill
joe@hoopnumbers.com
Improved NBA Adjusted +/- Using Regularizationand Out-of-Sample Testing
Joseph Sill
joe@hoopnumbers.com
