论文部分内容阅读
Growing up， Rachel Martywould spend hours shooting hoopson her driveway with her dad.
从小到大，瑞秋·马蒂会跟父亲一起在家门口的路上练习几个小时的投篮。
Sometimes， a talking computerjoined them. In the early years，Marty's dad connected a laptop to acamera， placed them on a cart， androlled it out onto the street's edgewhere he surrounded it with trafficcones. From about 25 feet away， thecamera could capture the arc of theball as Marty shot. And with eachshot， the computer would announcethe angle of the ball's trajectory.
有时，一台会说话的电脑会加入他们。早年间，马蒂的爸爸会把笔记本电脑与摄像机连接在一起，将这个设备装在手推车上，推到街上由交通锥组成的场地上。在7.62米外，摄像机能够捕捉马蒂投篮时篮球的弧度。每次投篮后，电脑会报出皮球行进路线的角度。
43.47.42.
43.47.42。
The optimal entry angle，Marty says， is 45 degrees， and thekey is consistency. With this hightechtrainer， she honed her shotover the years as a shooting guardin high school and in college at theUniversity of California， San Diego.
马蒂说，最理想的进球角度是45度，关键在于持续性。有了这个高科技训练师，她不断磨练自己的投篮，分别在高中和加州大学圣迭戈分校担任得分后卫。
"We were known in the neighborhoodas being the weirdos whohad this talking machine，" she said.
“邻居都说我们是跟机器说话的怪人。”她说。
But things have changed.In the last few years， high-techcameras have proliferated acrossthe highest levels of basketball. In2010， the year Marty graduatedfrom high school， a sports companycalled Stats， LLC， installed the firstSportVU camera systems in NBAarenas. These cameras， perchedup with the rafters， track not onlythe basketball， but also playermovement. By 2013， these cameraswere in every NBA arena.
但情況已经发生了改变。过去几年，高科技摄像机已经广泛应用于高水平的篮球赛。2010年，也就是马蒂从高中毕业那年，一家名为Stats的公司将首个SportVU摄像系统装进了NBA球馆。这些装在屋顶上的摄像机不仅追踪篮球，同样追踪球员的移动。到2013年，这些摄像机入驻了全部NBA球馆。
The use of machine learningand AI in basketball representsonly the latest chapter in theanalytics revolution that has beentransforming basketball overthe last 15 or so years. "It's just acontinuation of the spectrum，" saidBrian Kopp， a sports technologyexecutive who helped Stats， LLCroll out SportVU to the NBAduring the early part of the decade.
使用机器学习及人工智能只是过去15年内彻底改变篮球这项运动数据分析革命的最新篇章。“这只是这一领域的延续。”布莱恩·库普表示，作为体育科技工作人员，他协助St at s公司在早期将SportVU系统推广到了NBA。
Of course， teams had longbeen recording basic statistics suchas points， rebounds， and assists. Butwith analytics， as this statisticalapproach is called， the stat geeksestablished the power of math.
当然，球队多年来一直在记录诸如得分，篮板和助攻这样的基础数据。但有了数据分析后，当人们开始从数据角度看待一切，数据极客们确立了数学的威力。 That statistical sophisticationwas widespread in the NBA by thetime teams first adopted trackingcameras. But the new deluge of datademanded updated techniques andsoftware. "Along came SportVU，and the data we captured could notbe contained in Microsoft Excel，"said Kopp.
当球队第一次采用追踪摄像机时，NBA早已开始使用复杂的数据统计方法。但新型数据的出现，要求技术和软件也要做出更新。“这就出现了SportVU，而且我们采集的数据只靠Excel是没法承受的。”库普说道。
Meanwhile， increasinglypowerful computers were enablingnew advances in machine learning，so NBA teams and companies likeStats took advantage. The task wasto translate the tracking data intosomething searchable and digestible，and one major goal， in particular，was to identify specific actions likepasses and common plays such asthe pick-and-roll. Machine learningand， more broadly， AI were perfectfor the task because， they're aboutpattern recognition at their hearts.And pattern recognition is preciselywhat's required to distinguish， say，the many variants of a pick-and-roll.
与此同时，功能越来越强大的计算机让机器学习出现了新的突破，NBA球队及Stats这样的公司抓住了这个机会。他们的任务是将追踪数据翻译成可供研究，容易理解的信息;其中一个主要目标，就是明确如传球这样的特定动作，以及挡拆这样的常见战术。机器学习，或者更宽泛地说，人工智能，极为适合执行这个任务，因为归根结底，计算机的核心就是模式识别。举个例子，模式识别正是区分多种挡拆类型所需的技能。
To be clear， these algorithmsaren't doing anything humanscan't. Humans are really good atrecognizing patterns， and teamshave long hired staff to reviewvideo and identify notable clips forgame planning. But it's impracticalto sift through hundreds of hours offootage to identify every single pass，shot， rebound， cut， screen and roll.
需要明确的是，算法能做的事，人类都能做。人类其实很善于识别模式，多年来球队也有专门工作人员审查录像，确定值得注意的战术，从数百小时的录像中筛选一个传球，投篮，篮板，空切和挡拆。
Marty's talking computer --and steadily improved versions of it-- would follow her through highschool and college， helping to trainher teammates and even during herrecovery from injury. "The systemwas growing up with me，" she said.
马蒂会说话的电脑——以不断改善的版本——伴随她度过了高中和大学，帮助她训练隊友，甚至协助她从伤病中恢复。她说：“这套系统和我一起长大。”
In 2001， her father， Alan， hadturned his invention into a companycalled Noah Basketball， named sobecause in the Bible， Noah had theperfect “ark.” The newest version，which includes tools to analyze yourdata and is dubbed Noahlytics， isnow used by half of NBA teams，dozens of top college programs，and hundreds of individuals aroundthe country， said John Carter， thecompany's current CEO. Their systemconsists of cameras attached 10feet above the backboard that don'tjust track the ball's arc， but also itsdepth and left-right position as itapproaches the rim. The system collectsdata on every shot， adding to adatabase of more than 150 millionshots already recorded during practicesessions around the country. 2001年，她的父亲阿伦用自己的发明创始了名为“诺亚篮球”的公司;之所以使用这个名字，是因为圣经中诺亚拥有完美的“方舟”。据公司现任CEO约翰·卡特表示，被命名为“诺亚分析”的最新系统包括数据分析工具，如今已经被半数NBA球队，众多顶尖大学以及全美各地数百名个人用户使用。他们的系统由位于篮板上方3.05米的摄像机构成，这些摄像机不但追踪球的弧度，也会记录球的深度以及接近篮板时的左右位置。系统会收集每一次出手数据，补充到数据库中，这个数据库已经记录了全美各地超过1.5亿次的投篮数据。
This data is crucial for themachine learning algorithms thatare being incoroporated into thenewer systems. The next version，which has already been installedfor a few NBA teams， incorporatesfacial recognition， and can trackshots from multiple shooterssimultaneously. That allows thesystem to collect and analyze datafrom team practices， for example.According to Carter， in the springthe company will introduceanother new feature to capture thetype of shot a player takes. Usingdeep learning， a type of machinelearning that mimics how a brainlearns， the system can identify theparticular type of shot taken --moves that have varying degrees ofdifficulty that affect the likelihoodthat the ball goes in.
對于配备在新系统中的机器学习算法，这个数据至关重要。一些NBA球队已经开始使用有面部识别系统的新版了，它可以同时记录多名射手的投篮。举个例子，这个改变可以让系统收集球队训练中的数据并做出分析。据卡特表示，公司春天还会引入另一个新功能，捕捉球员的出手方式。使用深度学习（一种模仿人脑学习过程的机器学习），系统可以确定特定类型的出手方式——不同难度的出手，会影响进球概率。
But whether a shot is madeultimately depends on the ball'strajectory. Rachel Marty， whorecently earned a doctorate inbioinformatics， analyzed datafrom 22 million shots collectedby Noah's system to find a moreefficient way to evaluate a shooter'sskill. Typically， coaches and scoutsevaluate shooting ability based onhow many shots a player makesduring a workout. The more shotsa player takes， the better you canassess her true abilities.
可能否进球，最终取决于球的行进轨迹。最近刚获得生物讯息学博士学位的瑞秋·马蒂分析了诺亚系统收集的2200万次投篮，希望能找到更有效评估投手能力的方法。一般来说，教练和球探基于球员训练时的命中数量评估投篮能力。出手数量越多，就越能评估她的真实能力。
Since not every player has theopportunity to take thousands ofshots， Marty trained an algorithmto recognize the characteristics ofa made shot and predict a player'sskill based on only a few attempts.Instead of having players takethousands of shots， coaches andscouts can have them shoot， say，25. By analyzing the recordeddata， Marty's program can providean accurate estimate of their ability.This capability is also somethingthat will be eventually incorporatedinto their commercial product，Carter said.
考虑到不是所有球员都有机会出手几千次，25岁的马蒂表示，她尝试了一种算法，可以识别命中投篮的特点，且只需要几次出手就能预测球员的能力，教练和球探不再需要球员出手几千次。通过分析记录的数据，马蒂的系统可以准确估算球员的能力。卡特说，这个功能最终也会被融入他们的商用产品。 The impact of machinelearning and AI goes beyondX's and O's. Much of SecondSpectrum's efforts are to providea personalized game-watchingexperience with evermore detailedstatistics for the geekiest fans.For example， their CourtVisionaugmented video technologylike the probability a player wouldmake a basket， which is based onhis past data. You can watch thenumbers change above his head ashe runs around. The company's goalis for this technology to be widelyavailable on all broadcasts.
机器学习及人工智能的影响并不局限于赛场。Sec ondSpectrum公司的主要工作，就是为痴迷于数据的极客球迷，针对他们提供定制化的观赛体验。比如他们的CourtVision增强视频技术可以在球队过往数据的基础上，将球员的进球可能性等数据或图像叠加在视频中。球员跑动时，观众可以看到他头顶的数字变化。公司的目标是让这个技术广泛用于所有比赛直播。
随着体育接纳人工智能，机器最终会统治一切吗？不管怎么说，目前世界上最好的国际象棋，围棋选手都是电脑。未来NBA的比赛会降格到算法之间的竞争吗？在战略及战术层面，你无法像解决围棋比赛那样解决篮球比赛。复杂的棋牌运动需要遵守严格的规则，而篮球是流动的，球队需要不断做出调整与反调整。
An emerging market for thesereal-time probabilities is sportsgambling， which is now legal inmost states thanks to a SupremeCourt ruling in May that struckdown a ban on commercial sportsbetting. Much of sports gamblinginvolves wagering on specific eventsduring a game， such as whether acertain player will score 20 pointsin a quarter. To set the odds， gamblinghouses will need real-timeprobabilities. It might take a fewyears for this to develop， Kopp said，but the opportunity is there.
這种实时可能性的新兴应用市场，就是体育博彩，五月，美国最高法院撤销了对体育商业博彩的禁令，美国大多数州已允许体育博彩。很多体育博彩涉及在一场比赛的特定时间投注，比如某个球员能否在一节比赛里得到20分。为了确定赔率，博彩公司需要实时可能性。库普表示，这种技术可能需要数年时间才能发展完善，但机会就在眼前。
As sports continue to embraceAI， though， will machines soonreign supreme？ After all， the bestchess and go players in the worldare now computers. Would afuture NBA game be reduced to acompetition of algorithms？
随着体育接纳人工智能，机器会统治一切吗？不管怎么说，目前世界上最好的国际象棋，围棋选手都是电脑。未来NBA的比赛会降格到算法之间的竞争吗？
Unlikely， said Dean Oliver，vice president of data science atTruMedia， and considered one ofthe pioneers of basketball analytics.When it comes to strategy andtactics， you can't solve basketball theway you can solve a game like go.Even complex board games followrigid rules， while basketball is fluid，with teams constantly adjusting andreadjusting to each other. "There isno dominant strategy，" he said. "Thegame is far more robust.”
不过TruMedia的数据科学副总裁迪恩·奥利弗认为这种情况不可能发生，他被看作是篮球分析界的先锋。在战略及战术层面，你无法像解决围棋比赛那样解决篮球比赛。复杂的棋牌运动需要遵守严格的规则，而篮球是流动的，球队需要不断做出调整与反调整。“不存在统治性的战略。”他说，“篮球比赛更有活力。”
Indeed， a coach's job is farmore than drawing up plays. "Theelement of coaching that doesn't getplayed up as much as it should is theability of the coaches to work withplayers， to motivate them， to getthem to play together，" Oliver said. 事实上，教练的工作远不止指定戰术这么简单。“教练一项很少被提起的工作，就是与球员合作的能力，如何激励他们，让他们团结在一起。”奥利弗说。
Although the teams quickestto adopt machine learning do gainan edge， machine learning hasn'trevolutionized the game just yet， headded. "What's changed the gameof basketball from what it was 10years ago until now is not the AIand machine learning part of it，" hesaid. "It wouldn't be tremendouslysimplifying to say that StephCurry changed basketball." Curry’sdevastating ability to shoot fromsuch range has forced teams toabandon strategies that were onceconventional wisdom.
尽管最早接受机器学习的球队确实获得了竞争优势，但他补充，机器学习尚未给比赛带来革命性变化。“现在的篮球与10年前的变化不在于人工智能和机器学习参与到了其中。”他说，“极度简化的说法就是，史蒂芬·库里改变了篮球。”库里的远程出手能力迫使球队放弃了传统策略。
As a data scientist， Martydoesn't play basketball as muchanymore. But when she visits herparents' home， she tries to takethe time to roll out the old shottrackingsystem， just like thoseevenings she used to spend shootingwith her dad. "That was our specialfather-daughter thing，" she said.
作为数据科学家，马蒂不再像过去那样经常打篮球。但拜访父母时，她总会搬出旧的投篮追踪系统，就像当年她和父亲一起投篮的那些夜晚。“那是很特别的父女之间的联系。”她说。
Trying to hit that 45-degreeangle gets addictive， she said. But it'sfun， and sometimes it's always goodto go back to the basics -- even ifthat means a talking computer.
她说，以45度角命中投篮会让人上瘾。但这很有趣，而且回归基本是好事——尽管这意味着你需要一台会说话的电脑。