too small...
"A team source told SNY's Ralph Vacchiano that Oklahoma QB Kyler Murray is "probably a little too small" for the Giants.
The Giants "prefer to stick to the established measurables they have for a prototypical quarterback," Vacchiano notes. The Giants' organization emphasizes conventional wisdom and inside-the-box thinking, and they haven't started a quarterback that measures below 6-feet since 5-foot-11 Gary Wood went 0-6 in 1966. And "the philosophy hasn't changed all that much (in that over half-century timeframe)," Vacchiano confirms. 6-foot-3 Dwayne Haskins, 6-foot-4 Drew Lock, and 6-foot-5 Daniel Jones appear to be likelier candidates to succeed 6-foot-4 Eli Manning than 5-foot-9 Murray."
Source: SNY.com
Quote:
it’s a long term investment that everyone is going to have to makensooner or later.
To Christian’s point the data that exists for this now can predict these data systems but it doesn’t mean it’s easy to build them. By all accounts the Patriots have been building theirs for almost 20 years. The Eagles maybe longer.
This is a when not an if thing. And if you are just starting when basically many other teams have proven unequivocally that it works you are too late.
Did you read the article on chess centaurs I shared? Even if the Giants finally get with it and hire the right people they are still behind on the reinforcement learning data set and falling further behind even if they hired the best coders in the world they would be dealing with inferior data than those teams that have been using reinforcement learning for all this time.
Also I like that I’m weird.
Human salvation lies in the hands of the creatively maladjusted. ~ Martin Luther King
You clearly have very little understanding of reinforcement learning outside of this one article you read. There's too much open-ended complexity and a lack of a good simulation environment for it to be useful in the context of an NFL game. Data sets aren't the bottleneck...
Ummmm. I've been talking about needing to set up a good learning environment with good outcome analysis. The more time you have to train the agents and add to their complexity the better they will be.
And i'm not sure what you are talking about with lack of a good simulation environment? You can run simulations of past games. Feed in Zebra data, build an autoencoder like the Patriots have to deepen the longitude of training set with film. (among many other traditional data sets that you'd expect) So you have even 2 different data trees, film models, your Zebra + film vs. pure film models and you can also compare them against the human agent outcomes. Especially with the autoencoders built from film you build a pretty deep set to find the play calling agents which maximize win total and ultimately Superbowl victories. You can even build in structures to let them manage the cap / resource allocation (by creating a database of graded contracts $/performance) in the off-season in their pursuit of calling more successful plays maybe you want to separate them into funnels, probably you'd try both though. Ideally like in the article I shared you start showing the human "players" the recommendations and seeing what they decide on with that information, that's the holy grail data set that I am advocating we get to as soon as possible.
I also strongly believe that the best probabilistic models are built on top of the metadata on simulations from the reinforcement learning agents, as well as unsupervised and supervised methodologies if we are getting greedy. These goal state optomizers can even feed into larger goal state optimizations. The success outcome and resource allocation could feed into fan sentiment and pricing strategy to feed into profit.
I actually have filed two patents and am about to get my third jointly with a large university with a prominent professor in one of the top applied math schools in the world. The new one is on mass signal interpretation and model verification. The ones we've filed, one of them is for an original approach to multi-agent frameworks and goal state orientation and the earlier one for automated feature engineering, hyper-parameter tuning and an evolving neural network feedback loop. We published on ENNs two years before Sentient wrote about them.
Quote:
In comment 14297458 Ten Ton Hammer said:
Quote:
on how to argue something in a completely unlikeable fashion that drives people away from supporting your point.
What happens when you argue this point in 4 different threads with the same person and he keeps on acting like he has a point when he doesn't? Do you do it the same way again? Sounds like the definition of insanity. Have to try troll hunting in different ways
Here is the difference between some of the people arguing my side (that the Giants need to improve their tech commitment yesterday) and the luddites. The luddites are arguing with very limited facts and information supporting their side to needlessly defend a regime that is already in place. The Giants are already doing this with their support or not.
The other side would like to see change in the organization that could really do a lot to help us field more competitive teams going forward and it would certainly help if more fans acknowledged this was a problem instead of pretending it wasn't.
Owners talk to each other. I'd also think by now he isn't going to forgo a tool that is very likely to give his team an advantage.
They DO use analytics, at least nominally. The issue, IMO, is whether they know what questions they're using analytics to answer, and whether they're dedicating enough resources to do so effectively. We'll probably never know about the former - that's proprietary after all - but to the degree that org charts are even semi-public, the Giants' analytics team does seem a bit thin on headcount.
The other side would like to see change in the organization that could really do a lot to help us field more competitive teams going forward and it would certainly help if more fans acknowledged this was a problem instead of pretending it wasn't.
How will fans acknowledging this issue help?
Quote:
The other side would like to see change in the organization that could really do a lot to help us field more competitive teams going forward and it would certainly help if more fans acknowledged this was a problem instead of pretending it wasn't.
How will fans acknowledging this issue help?
The Mara's certainly have a pulse on the fans, it's important to them. The more that it becomes accepted around the fan base that this is something that needs to be addressed the more they hear about it (as a pure theoretical numbers game) the higher their headline risk of not addressing it and the higher chance they will.
Again even if they just bring in a strong CTO type that is in the draft war room, interjecting about what the models are saying, has a real voice. That's the start.
The early value a strong predictive analytics program isn't even answering the tough questions, it's interjecting a different perspective that might re-frame debates. When the debates are re-framed and the shortcomings of the models called out, you figure out what people are seeing that the models aren't find a way to encode that information and it becomes a virtuous cycle of improvement.
The problem you've run into multiple times is NOT what you suggest here. Many people (including myself) can agree that perhaps the Giants can do better in analytics.
The problem is that you have made assertions to the positive about the level of analytics the Giants have made as though you actually know it. You've used "evidence" that includes a single line stated and possibly taken out of context from a press conference, what you could find on Google, what sportswriters say, and what people's LinkedIn profiles say about them (not even their actual resumes).
You have stated that this is ample evidence to support your conclusions. Many people disagree with you on whether that is ample or evidence.
To further complicate your situation, you've taken to labeling those who disagree with you as luddites, which is an absolute mischaracterization of their viewpoints and arguments.
Let’s start with talent identification & recruitment. Here, I believe the Giants should have a capable front office analytics dept…I think analytics is a great way of challenging preconceived notions and group think. I would absolutely use it for that. I would not use it as the primary input to decision-making. My scouts would be key. Here are my concerns with analytics in talent identification…It doesn’t reflect chemistry, leadership or potential well, the sample sizes are much smaller than baseball, there’s so much data you can generally torture it into confessing what you want (copyright Bill2!). I also question if there’s value in blazing the trail or being a follower. It will take a lot of costly mistakes to get the learnings…especially on things like machine learning that are very immature. Think about what goes into analyzing one CB on one play. What’s his leverage? What’s his cushion? What was the scheme calling for? How much safety help did he get? What was the quality of the throw? The down and distance> Things the human brain processes very swiftly but are actually very tricky to code for reliably and require a ton of data points to cover.
Regarding strategy, time management and health management I think the benefits are a lot clearer. The data is much more easily measured, more easily applied and should be basics but let’s be honest, there’s no silver bullet to these things because no-one wants to put the maths guys in charge. Andy Reid is one of the biggest advocates for analytics in the NFL. He is also a famously bad caller of timeouts. Atlanta are huge proponents of analytics, yet Mike Smith was allowed to be the worst user of timeouts the league has arguably ever seen. Let’s not cover them losing a superbowl where basically they could have knelt on every play at a certain point. ARC does a great job discussing the Jags. So I think my point here is that there’s a bunch of value in doing the work but only if you put it in to action effectively, which there is some evidence Shurmur is doing in a limited range. However, let’s be honest, his timeouts were atrocious in 2018, especially the first half.
Overall then I’d characterize my approach to analytics as someone prepared to embrace it very much in some areas and cautiously in others where there are real barriers to implementation and risks of costly learnings. If someone wants to actually lay out how we mitigate those concerns, get the bang for the buck without impacting the available time of the key guys during the season then I’m all ears. Can we just try to keep our responses to less than 3 volumes and a teeny bit sane. Thanks in advance.
Generally speaking I agree with thispoint of view.
I will point out that Machine Learning is by no means immature. I was involved with machine learning techniques and algorithms back in the 80s. From a computer science perspective, the techniques and algorithms are quite mature. It has only come to the current forefront in the past decade or so because we now have the computing power to make those algorithms worthwhile.
That said, the dichotomy you draw between player health/peak performance& game management (the easier stuff) vs. player evaluation & recruitment (the harder stuff) is accurate. For sure get the easy stuff going first, that's table stakes. The thing about the harder stuff though is that the benefits are not always obvious. The benefits are about helping to make higher percentage decisions. No decision, no matter how much data reinforces it is 100% guaranteed to work.
I don't know if the decision to draft Bortles was driven by any analytics or just desperation to get a QB. And if there were analytics behind, were the right variable being analyzed? No idea. Even if there were analytics on the right variables, you only get an indicator that says you have an (x)% better chance of building a superbowl contender with this pick. The pick can still go wrong. Frankly, I don't see an analytics based approach that would lead to the Bortles pick, I think there was enough in the data to steer away from that one, but what do I know.
Using one failure to discredit an analytics approach is incorrect thinking about it. Any tool that helps you to make the higher percentage decisions is a good tool. Not using a tool that improves your decision making is straight up a bad thing.
Given the revenues of a football team, the cost of building up a decent analytics department is table scraps... There is no reason not to do it. And the sooner you start, the sooner you start building up Intellectual Property on what works and what doesn't. That IP can take years to build up, and the longer you wait, the harder it is to catch up. Other teams are not going to share their IP, the IP is not doing to land in the public domain since it is so highly specialized. It comes down to a black and white decision, only limited use of analytics for the easy stuff and reject it for the harder stuff, or jump in with both feet and do both.
What fans say the Giants reject analytics, what they are really saying is that the Giants are rejecting analytics for the harder stuff. Everybody is doing the easy stuff, so that gets discounted. Fans who are critical of the Giants decision making and know something about computer science and data analysis are frustrated because we know their is a tool out there that can help with those decisions, and that tool is being rejected for specifically those decisions.
Quote:
Think we got off on the wrong foot when more than a few people called me stupid for suggesting it was possible for the Giants to do better with this. I will definitely have a response for you soon. For my part I apologize for being too nasty about this on other threads.
The problem you've run into multiple times is NOT what you suggest here. Many people (including myself) can agree that perhaps the Giants can do better in analytics.
The problem is that you have made assertions to the positive about the level of analytics the Giants have made as though you actually know it. You've used "evidence" that includes a single line stated and possibly taken out of context from a press conference, what you could find on Google, what sportswriters say, and what people's LinkedIn profiles say about them (not even their actual resumes).
You have stated that this is ample evidence to support your conclusions. Many people disagree with you on whether that is ample or evidence.
To further complicate your situation, you've taken to labeling those who disagree with you as luddites, which is an absolute mischaracterization of their viewpoints and arguments.
No there is much more evidence as to what the Giants are using analytics for. THere are articles where the team has said they use it for peak performance and health management. I think it is safe to assume that they use it from some in game management like when to go for it on 4th or when to go for 2...
Teams that are using it for player evaluation and talent acquisition have much larger departments than the Giants have. All the evidence out there shows that the Giants have 1 guy who has a specialty in the health industry doing their analytics (which goes in line with the fact they use analytics for health and peak performance). When researching other teams, its easy to find out that they have larger departments although the size is not readily available it is referred to as "teams" of computer professionals. And the leaders of these teams have strong backgrounds in field such as computer science, machine learning, big data, statistical analysis and often investment analysis. The Giants have none of that. Without such skills it would be impossible for the team to delve into the player evaluation & talent acquisition side of analytics.
The on field results have been poor. And there has been zero change in our analytics personnel so it is hard to argue the competence of our staff on that note.
Regardless, the timeout and time management blunders in conjunction with the evidence we found is what makes some of us believe it's a strong argument.
You can post articles about Gettleman understanding advanced game theory as much as you want but when we fail to see simple game theory applied in the games we watch it certainly makes the argument that we aren't committing enough or the right resources stronger than the converse.
When the results on the field are so poor it is incredibly irritating for people to ignore this other evidence. Even the Jags their fan base can say that there is a commitment towards building technology for the future even if the results aren't there now.
Do you really think it is too much to ask for either good performance on the field or a clear commitment to evolving the approach in a rapidly changing tech environment?
Honestly my fear is that they fix the simple game theory stuff this season and never address the team construction things and we find ourselves badly behind on that because voices that are so loud now without any evidence on the low hanging fruit that shows we have the right tech people.
I guess my last question is why is it such a problem to want a company in 2019 that makes billions to have a plan as it relates to technology with more visibility? Why do the people suggesting this deserve ire?
That's what I've been reacting to, my increasing frustration that those of us that want this need to be called stupid or lambasted generally.
NoGainDayne said:
The biggest thing that we've gone round and round about relates to determining the disposition of the Giants towards analytics. An assertion has been made that they are NOT supportive of analytics. Over time it appears that assertion has softened, but I think what remains is an attitude of certainty about how the Giants are currently approaching the use of data in all of their football operations.
For someone like me it is puzzling how you can say with certainty that there has been zero change. First of all, it is most likely (imo) that the Giants recognized their limitations in this area and determined (as has been their M.O. in the past) to contract OUTSIDE the organization with people who are best qualified to help.
How would any of us possibly know if they are using the same or different people/methods than they were in years past? It seems really out there to suggest WITH CERTAINTY that zero changes have been made.
Furthermore, the analysis of abilities based on LinkedIn seems ludicrous to me. If you checked my LinkedIn profile you would see NOTHING that I've been working on for the past 6-8 years. My skills have completely evolved and yet I prefer NOT to update my resume or profile. Why? I already have the job I want.
While it may be very likely that those named by the Giants as overseeing analytics are not skilled enough to start up and run an intensive analytics department, it is very possible that the skills needed by the front office in terms of understanding the findings/analysis of said department and communicating it well to those who make decisions. Truth be told, a skill that is highly prized (and needed) is the ability to tactfully navigate between those who do deep analytical work and those who execute in operations. There seems to be ample evidence that not all people who are at the forefront of this technology are excellent diplomats in relating to those who run a football team.
You can post articles about Gettleman understanding advanced game theory as much as you want but when we fail to see simple game theory applied in the games we watch it certainly makes the argument that we aren't committing enough or the right resources stronger than the converse.
You are correct that poor results are strong evidence that things are working, but off the mark (imo) to use such an incredibly small sample set to come to that conclusion. IMO newly hired football leaders need time to reverse course when it comes to how the team performs. It is very possible that DG and all the front office is unhappy with PS's performance and that if it continues, there will be a change. Isn't that just as likely a conclusion as one that suggests the Giants have no ability to use game theory to develop strategies in such simple situations?
Do you really think it is too much to ask for either good performance on the field or a clear commitment to evolving the approach in a rapidly changing tech environment?
Honestly my fear is that they fix the simple game theory stuff this season and never address the team construction things and we find ourselves badly behind on that because voices that are so loud now without any evidence on the low hanging fruit that shows we have the right tech people.
It's one thing to be frustrated with results (we all are) or to have a fear of something. I think most Giants fans have similar if not identical fears. It's entirely another to draw conclusions about cause and promote them as proven based on evidence when that evidence is not clear to everyone.
Heck, if you took the approach from the start that all of your assertions were your opinion instead of presenting them as evidence-backed facts, I never would have involved myself in these discussions. I am not in a position to argue against analytics, nor am I in a position to provide evidence to support any claims about the level of investment the Giants have. I am super-curious about it though, so I took the time to read through a HUGE thread where you claimed to have left ample evidence and didn't find anything that rose to the level I would allow a HS student to use to back a claim.
It's not a problem to want the Giants to have a plan for tech. Not sure I agree that it should have a high-level of visibility. Maybe because I'm not trying to make a living in that field?
I don't think anyone who wants this deserves ire either.
This is why I asked you yesterday to focus on FMiC's main point - it seems you look beyond the idea that what is evidence enough for you is NOT evidence enough for everyone else, and instead turn this into something it hasn't been - an argument for or against analytics.
Well done.
1) I think this is just where we believe different things. For me not having a strong technology leader, having a leader speak out against analytics, your "rising star" not having advanced math or software experience (and yes people can learn things but it's definitely pretty hard in a full time job and Sean Harrington for example probably had what looks to be 7 years of computer science and software engineering work head start on Ty Siam), poor applied game theory demonstration, advocating secrecy (I've explained on this thread why this is a backwards way of thinking), other teams having good press and public leadership embracing analytics. That to me constitutes a negligent attitude a non-supportive attitude in general of the types of inroads you need to make compete in a large and quickly changing technological industry. We just disagree on this, I feel others with knowledge of analytics other than you agree with me but I don't think this position should have ever been called stupid.
2) I just really disagree that you should have time to get the timeout thing right. I think Christian spoke very well on this above. Not hard to hire someone whose job it is to talk to Shurmur about these probabilities. The Eagles clearly do it for much more complex things. Fact is the talent in New York is wildly better than Philly or many other cities to this should be even less of an excuse.
3) When I said zero change I meant zero public change. We've outlined the merits of having internal vs. external people and I think the points I made about needing to build our own data sets are extremely salient. If you hire outside talent you not only essentially making the intellectual capital of your team available to others for purchase but you run the risk of being extorted for money or sent back to the stone age.
4) I think your burden of proof is unrealistically high and while we are throwing insults like HS student around. There were more than a few people who appreciated the evidence and research I put into that post. If you are so confident that what you are doing is better you aren't afraid to put out old stuff to pull in talent. Every successful tech enterprise does this including the Patriots. Nobody is asking the Giants to put their code in a public git. If you aren't talking it is way, way more likely that you don't have anything to talk about or people want to research than you are ahead of the curve. Not wanting to talk about what you do screams of a fear that you might have only had "one good idea" people invest in you, people hire you, etc. if they believe you have had ideas that other people don't and will continue to do so.
Not only do the folks in the software development and tech industry look at linked in. There are other places that probably carry more weight. For example, I have found the GitHub repositories of various people working on analytics for other teams. GitHub is a place where people can manage their source code. Things that are not protected IP can made publicly available. Companies today will often look to an individual's GitHub repo for evidence of their prowess with a specific tech or sub field. Furthermore they look for a continuous history, not just recent updates, so not keeping GitHub up to date really hurts candidates unlike LinkedIn. On GitHub people may not say exactly what they are working on, but they will usually say for whom they are working. Contractors will name the client company. There is nobody that I can find that has a GitHub repo that mentions the Giants. There are literally hundreds of mentions of other teams. Its easy to find people who've worked for the Patriots, Browns, Panthers, Eagles and Chiefs. In fact, the vast majority of teams can be found referenced in GitHub... The only mention of Giants is the SF Giants, not the NY Giants.
If there were contractors doing this sort of thing for the Giants, reporters would ferret it out. Perhaps not the details, but that it does exist.
Furthermore, the idea of using contractors or outside experts for anything involving true IP is a very BAD idea. No team would want the knowledge gained by working on their proprietary models to be taken to another team. Any agency worth their salt would market this expertise to other teams if they had it. SO I believe that it is highly unlikely that ANY team will outsource any of their models, algorithms or data analysis.
For someone like me it is puzzling how you can say with certainty that there has been zero change. First of all, it is most likely (imo) that the Giants recognized their limitations in this area and determined (as has been their M.O. in the past) to contract OUTSIDE the organization with people who are best qualified to help.
How would any of us possibly know if they are using the same or different people/methods than they were in years past? It seems really out there to suggest WITH CERTAINTY that zero changes have been made.
Furthermore, the analysis of abilities based on LinkedIn seems ludicrous to me. If you checked my LinkedIn profile you would see NOTHING that I've been working on for the past 6-8 years. My skills have completely evolved and yet I prefer NOT to update my resume or profile. Why? I already have the job I want.
While it may be very likely that those named by the Giants as overseeing analytics are not skilled enough to start up and run an intensive analytics department, it is very possible that the skills needed by the front office in terms of understanding the findings/analysis of said department and communicating it well to those who make decisions. Truth be told, a skill that is highly prized (and needed) is the ability to tactfully navigate between those who do deep analytical work and those who execute in operations. There seems to be ample evidence that not all people who are at the forefront of this technology are excellent diplomats in relating to those who run a football team.
Dorgan - you are in the coaching field if I remember correctly yes? I would think you have a deep connection to the traditions. Understandable. And there seems to be a bunch of pitchforks out with people being like. We want proof, proof! But not much evidence supporting your cause other than pointing out that we might not know the absolute complete picture about what is going on behind closed doors.
Instead of fighting us might it be nice that we all agree that it would be nice to get these answers from the team? I don't think people are grasping that I would be very happy to be wrong here, if there was a secret skunksworks of ex google engineers working in the basement. It's just very unlikely and hardly a reason to assume that maybe it is time that fans agree that this is addressed publicly in some way. I just think it is very very hard to hide that stuff. How do you keep those interviews quiet? Are these people not allowed to leave the building? Talk to their friends?
It seems like many would really just prefer to see a world that we are in good hands despite our franchise looking objectively pretty bad right now.
I do have many years experience working with technical contractors. Among other things, I was a project manager for a large U.S. bank (the largest in the U.S. at the time) on a major technical initiative which won me a major company award (awarded once a year to a single employee out of >120,000). I know about what gets shared publicly and how to create contracts (and often more importantly relationships) that ensure silence. The supposition reporters can and will ferret out the truth is acceptable to me, given enough time and interest in the topic. I expect that some day the truth will be revealed and it may prove to be very embarrassing for the Giants. To this point however, reporters have been shooting blanks at the topic.
I will say that the comments on your findings on GitHub are very, very interesting (far more interesting than what was presented earlier as ample evidence) and significantly strengthen the claims made, but still (imo) cannot be considered conclusive.
There is evidence that has been presented before (which I am unwilling to research again) that contract work is being done for NFL franchises. I believe that is far more likely than the alternative narrative - that the Giants eschew any tech advantages and are stuck in the realm of spreadsheets and the like.
Bottom line is this: The arguments I've made both past and present relate to how much evidence is necessary for one to make significant claims. I see no point or need to debate it further. If you are hurt by how people react to you may I suggest you temper your assertions in the future by stating your opinions less as fact and more as insights, as qualified as they may be.
Peace to you - may the Giants figure this thing out so we can start winning championships again.
I do value civility when it is afforded to me.
Bottom line is this: The arguments I've made both past and present relate to how much evidence is necessary for one to make significant claims. I see no point or need to debate it further. If you are hurt by how people react to you may I suggest you temper your assertions in the future by stating your opinions less as fact and more as insights, as qualified as they may be.
I have certainly never complained, nor do I feel hurt, about how anybody has treated me on this particular subject. Neither have I stated unequivocally that the Giants eschew analytics. Quite the contrary, it is my opinion that they do use them but in a limited capacity. We also know that every team buys data, that is certainly worthwhile outsourcing. No need to collect that data yourself when it can be bought for cheaper. However, no team in the league is going to want their proprietary models escaping, those will be closely guarded in house just like every investment company guards theirs. Outside contractors get nowhere near that IP. As those models mature, teams will realize that they need specialized data points to further their predictive capabilities. This will likely be data that is not readily available and will have to be gathered by the team. Again, this is not an activity that the team would want to be outsourced. They don't want other teams to know what data points they deem important.
Given that you work generally in tech for the financial industry, these should be concepts with which you are familiar.
The fact that the team's list of employees (which is public information, though not everybody's function is public) does not include anybody with an appropriate background is pretty strong evidence in my opinion. FYI, in another thread a few weeks ago, I and a few others went through the list of employees that had no clear function (only about 10 or so people) and searched out their work history (most turned out to be secretaries, some no info was available).
I have never once in the engineering, program, product, or applied science level ever heard of or known anyone who has worked on a project for the New York Giants. The Giants have never presented at or showcased to an industry group I've been associated with, and have never been mentioned in any peer reviewed or referenced study. I've never known anyone recruited to or approached by an agency or vendor associated with the Giants.
Of course that's not to say they don't have a solid team. Being a huge Giants fan and knowing literally thousands of people in the industry and having hired teams and vetted many agencies in the field, I suspect I would have heard something. Maybe not.
But again, even if the Giants have a team of a hundred people, they suck at the basics.
I have never once in the engineering, program, product, or applied science level ever heard of or known anyone who has worked on a project for the New York Giants. The Giants have never presented at or showcased to an industry group I've been associated with, and have never been mentioned in any peer reviewed or referenced study. I've never known anyone recruited to or approached by an agency or vendor associated with the Giants.
Of course that's not to say they don't have a solid team. Being a huge Giants fan and knowing literally thousands of people in the industry and having hired teams and vetted many agencies in the field, I suspect I would have heard something. Maybe not.
But again, even if the Giants have a team of a hundred people, they suck at the basics.
+1
LOL
C'mon man. It still hasn't sunk in over 4 threads what my point was, has it? My point was very simple.
Gettleman doesn't shun analytics and implemented the department in Carolina. Full Stop.
It is why McL brings value to these discussions - he's not arguing that point. He's not being an ignorant cock.
He's claiming that the Giants may not be on the cutting edge of analytics, or which I don't think anyone is denying.
Let me say this very plainly since you think I speak in platitudes. Gettleman doesn't shun analytics and implemented a system in Carolina. Both points of which you claimed weren't true. That's it. I have not argued any other point on any of those 4 threads.
And your response? Numerous references to Luddites and eating farts. Spectacularly moronic.
- Gettleman literally mocking the value of analytics publicly on a decision where it could really help
- The lack of presence of really any of the skills needed to build advanced analytical models in house
- No press on the Giants commitment or efficacy in analytics in the press and much more of this information on many other teams
- Most importantly the lack of seeing simple applied game theory on the field for the Giants (which is what brought me to speaking about this in the first place)
Your evidence
- One article which the above points really call into question
I started talking about farts and making fun of you in extreme ways because you have this seemingly steadfast loyalty to DG and the Giants despite very little evidence supporting your points.
What's more, you started calling me stupid assuming I didn't have any basis for my points before you even dug into my reasoning. You make fun of me for listing my credentials but a lot of that came out of you suggesting how could I possibly understand things that the GIANTS don't.
You get on me for jumping to conclusions but it is you who came at me with a lot of vigor and your bullying tendencies and I reacted by treating you with the same level of respect, somehow that makes me "pompous" and an asshole, funny coming form you a notorious asshole who just assumed I was coming at this uninformed when I was very informed and have read about this for years including the article you've harped on time and time again.
I never contested whether or not he hired engineers or went to the sloan conference, I contested that DG or the Giants have the kind of people that actually understand what it is possible with analytics and how to identify and bring in the RIGHT people to build these systems.
I like Christian's line about them potentially having vans of PhD's in the parking lot and still getting easy table stakes analytics / game theory things wrong. You can quote that article a billion times about Gettleman using advanced game theory but it isn't unreasonable for fans like me that understand game theory to say it doesn't matter what that article says.
There isn't proof but there is just a lot more evidence that Gettleman doesn't understand the capabilities or how to use / integrate analytics properly than he does.
I can understand a position where one could say "no that is not what I mean by applied analytics". I can understand saying that is health management and is separate.
My guess is that for you, that is not considered to be applied analytics. correct?
One thing I might add, the process the Giants have regarding health management/peak performance started under Coughlin, somewhere around 2008. So Gettleman would have been here at the time, and probably saw some value in it. When he got to Carolina, I am guessing he instituted something similar.
What I'd say about this is I have interacted with Deloitte or other top tier consultants most days for the past 20 years. I've managed serious spend on these guys (tens of millions a year for top investment banks and commodity companies). As with any industry, there's a huge variety amongst the resource pool from moderately capable in a narrow range to absolute rock stars. It is really difficult to spot the latter just from the CVs but what I find is those good ones are highly versatile, learn new skills incredibly quickly and hugely motivated to deliver.
Let's be clear though that no-one becomes an analytics expert overnight and for sure we need to be clear that machine learning, mathematical modelling or AI skills are very different and rare skills that I agree the Giants don't seem to have in house.
So should they? In the medium term I think so. Absolutely. Was it the most pressing problem Gettleman faced? Nope. He had to overhaul the scouting dept and coaching staff, get on the same page as his coach and fix a uniquely flawed strategy of fielding a statue QB with a terrible offensive line. So I can't fault DG for prioritizing.
In his shoes I would have kept analytics firmly on the back burner and done the basics (physiological data etc) and parked the hardcore stuff hoping that someone actually does a lot of the work for you and brings something resembling a mature, standard package to the market you can build off with some of the early adopter errors ironed out.
So I think for us to close the gap between us let's try and agree on the following.
1. Analytics can absolutely add value, across a range of different areas and we should attempt to catch up vs the rest of the league but it is trickier to apply than say baseball & basketball.
2. Some of those areas are easier and more mature than others. Even teams embracing analytics are doing stupid shit like drafting a non-generational running back #4 overall (looking at you Jacksonville) or managing the clock worse than your average Madden player (cough Andy Reid).
3. There are massive demands on people's time turning this ship around, scouting, investigating human beings, self-scouting, aligning the football strategy.
4. There is finite money
5. A poor analytics implementation is likely to deliver very little value. When it is done it should be done properly and maybe year 2 of DG's reign after the draft is a sensible time to kick this off.
Let me close with what I see as the value in analytics and why we shouldn't obsess with the how or the maths or the machine learning but with the fact that our team is thinking analytically.
I've forgotten some of the details of this story but if I remember it was the 1987 Giants - Redskins game in the playoffs for a superbowl trip. Belichek stays awake watching film for crazy amounts of hours with his staff and one of the quality control guys spots that (I think) Ricky Sanders doesn't put his mouthpiece in when he's not a route option in the playcall. The Giants then know they can jump other routes and amend the football strategy accordingly. We all know the outcome.
That to me is just a good example of the kind of stuff we should be looking to analyse automatically and teach machines to do for us. It does get done today though, quality control guys are obsessive about looking for these give aways and tendencies but wouldn't it make life easier to be able to see percentages etc as soon as the all-22 film is available and have a machine evaluate all of this in a few hours and you can constantly build up the scope.
Belichek may be ahead of the field now, he always has been.
Building the System Step by Step
1) A Play Type Forecaster – timeouts/time management with game theory. Feed in game scores and timestamp every play with amount of time left, quarter, down, distance and then start with something I’d say as simple as timeout, run offensive play, FB, punt, run defensive play, FG block, punt block, punt return then 2-point conversion as well. Each of these will have a win probability associated with choosing that type in a vacuum.
2) Enhanced Situational Play Type Forecaster – 4th down conversions and improved efficacy of step 1 systems through increased sophistication from data on weather, players, match-ups, team stats.
3) Build Autoencoder – this would convert game film to machine learning signals, I think you could look at this is a very similar to Zebra data, but this is important for the longitude of training sets as video goes back much further than alternative data
4) Build initial models for complex problems - stats, injury, asset allocation, play suggestions and fan sentiment could all be built and provide some level of value at this point (I’d use a combination of unstructured and reinforcement learning as well as an experimental framework for conditional ensembling)
5) Lay groundwork for meta models which combine inputs from step 4 like wins and profit
This could all be done in one year, step one a few months max, at the total cost of $.5M to $1M, now I am saying this with proprietary software I’ve built over that past almost decade, I’m not the only one with software like this but there are many pretenders too. I also think the spend that gets to 10’s of millions is essentially paying for highly paid data scientists to tinker with models, hyper-parameters, data structures and time series modifications when software is much better at doing this. I’d also add that McL who is clearly knowledgeable is talking about building much bigger teams than I think are necessary. I would also keep the team to pure back end until results passed the eye test and software deployment became a much bigger need.
I think the point on how valuable on this? Do you prioritize this? It’s tough to draw clear conclusions. In the case of the Saints game, down 12-7 at home completely blowing the creation of an extra possession for ourselves as we sat on timeouts when they had 1st and goal and more than enough time to score. (This wasn’t our only blunder on clock management on the season, but it was our worst) can we really prioritize fixing other problems over something like that? Winning that game could have completely changed our season. Can we say definitively that the way we are approaching “fixing these problems” are right when the larger focus on traditional methods while de-prioritizing advanced team building analytics has seen us fall to the bottom of the league? And even last year with new management supposedly with skills assessing the trenches have us hand out some bad contracts in FA to the positions they needed to fix the most?
I do think it’s hard to find the right tech people but that isn’t an excuse not to start spending what is a paltry amount relative to the profitability of the organization on trying. Contract out a search team, tell them they can't interact with Gettleman or his staff for the first year or two if you are that concerned about his time. Resources are finite but I do believe the main thing that needs to be “fixed” is that no matter how good our football people are now other teams have more diverse data to inform their decisions. While this isn’t a definitive theory it can’t be disregarded that a lot of these things we are trying to fix by largely the same methods could very well be symptom of the same problem which has to be fixed with the same commitment to technology that other teams have made already.
Wins (controlled for strength of team, not difficult if you have a way to project every player with an autoencoder)
Comebacks (controlled for strength of team)
Interview teammates and coaches and analyze with natural language processing (NLP)
Ask them about leadership, have them write about leadership (NLP)
Analyze grades, classes
Analyze facial expressions, body language and that of teammates
Wonderlic, existing Q&A etc.
This is the worry for me, something like this is just a jumping off point and based on Ty Siam’s LinkedIn there is a very low probability he can build systems that simply predict IT factor leveraging this data let alone nesting this solution within broader systems.
And this is also why the sooner you start the better, maybe you are not videotaping prospect interviews, maybe with players permission you want to start getting EKG or MRI readings while they speak to you. The most important thing is taking your first shot and figuring out what people might be capturing that the computers are not and that is purely iterative.
What’s crazy about this, is this a random thought exercise based on what we perceive to be something you need humans for when it’s possible that all this information is embedded in camera or Zebra data and you don’t need to do this at all. There are most likely a wealth of other shortcomings that can be addressed. That’s why people pay people like me to do what we do, you must know how to start as unstructured as possible and commit to a process of adding structures and data to attempt to improve and solve perceived shortcomings of the human end users. This can take up zero time of existing internal people, one dedicated scout could probably do the trick on this and you could hire them separately.
When it is ready to start rolling this stuff out to higher level people to start comparing to their decisions to the computer live for better reinforcement learning it may seem time consuming but it’s the job of a good technology leader that by the time they put it in front of business people with limited time it is already in a place to teach them new ways to look at the problem, so the models are improving as well as the perspective of the decision makers.
The scale is completely different, much smaller, for a football team.
Nowadays you don't need a datacenter, system admins, or DBAs, you can get all that from Amazon or Google or Oracle...
But you still need somebody to design your database, manager your data, build and maintain stuff to load the 2 dozen or so purchased data feeds that are purchased, load scouting reports. You need somebody to create models and analyze the results. You need somebody to do the software development of the model. As it grows and benefits are seen, more questions will come and the needs will spread out like tentacles. It won't take long before each branch has a person or two assigned to it. Before you know it you have 20 or so people. You will start off with maybe 5 or so, and maybe it takes 12 or 18 months for that to grow like that, but it will...
The other major point I want to make. I thought about EA's point about DG having higher priority things to do... However once last years draft was over, that was the time to start this process. What I was trying to say earlier is that if you hire the right person and give the a remit and freedom to pursue it, it doesn't take much of DG's time. DG won't understand what's happening anyway. He just has to hire somebody and say GO!
Do you realize how moronic that take is? That people are intentionally tanking on technology to discredit it?
And that's basically the rope you're hanging yourself with. You've already stood by the stance Gettleman shuns analytics, and once presented with facts that he doesn't, you now have to take a completely ridiculous angle that he likely hired bad tech guys.
Look man, the training room in both Carolina AND the Giants have been tracking nutritional data and using GPS data for understanding fatigue and efficiency of player movement for several years. There are many layers to what is being tracked. the debate should center around whether or not the data is being used correctly - not that it is being done incompetently on purpose.
But cue more nonsense about luddites and eating farts.
For all the bluster about people not makming points, your rants on luddites and eating farts is some of the most stupid shit posted on BBI. And that's fucking hard to do.
Nutrition and workout data is basically the absolute minimum required to check the box of "has an analytics department." I have no idea how anyone can use that to laud Gettleman or to suggest that he's actually embracing analytics. Especially when he very publicly mocked people using a data-based approach to evaluating talent and roster construction.
And I said I wonder. Here you go mis-classifying things very straightforward things I said again. I can't believe you are still calling me a moron. You are the one that has found yourself in the center of these debates without much knowledge of analytics.
I got out of the automated machine learning game about a year ago because of programs like Google's Auto ML and C3. I know one of the primary developers of their platform and the projects they do and the scale of what it can do is truely a wow moment. I did a short code walk and the magnitude of what they have put in modules of the data science process is staggering. But to speak in specifics about what you are talking about hiring for and what I know factually can be done in an applied finance setting.
Design a database - we have automated a lot of this and have admin software to verify the computer has done this correctly. We read in info on columns to figure out if they are strings, what kind of decimal to store numbers as etc. Most primary keys we can identify just by reading in the relevant data sets unless you go across data providers but even then its one a one off. Anything from an API or CSV we just read into Google Big Query and then store model / feature / ensemble metadata in a cloud database which makes heavy lifting requests to big query based on what is being used to solve the optimization problem or goal state.
Manage Data - We have a fairly exhaustive feature engineering system where we create many different kinds of versions of a column like a moving average, % difference etc. and our newest tech that I can't go into because it can actually explore an infinite number of time series intelligibly. Also it's pretty easy to just run anomaly detection like % 0, % null and flags for extreme deltas in summary stats which we both use for this purpose and as binary variables in our learning.
Models and Results Analysis - We use about 20 different algorithms and have them explore within logical hyper parameter plains. Sometimes we just constrain training times if our results are super rushed but otherwise certain numbers of layers, trees, etc. I always monitor what is being built, we also have ways to ensure diversity not just sheer numbers. Depending on what we are trying to solve or optimize for we will layer information coalescing systems on top of this base.
This is where we use most of our human labor and why I'm always talking about building your own data in reference to this problem, using this metadata creation and experimentation machine to figure out what parts of the problem we aren't grasping and then try to turn qualitative approach into additional data or module structures that can feed into the problems we are trying to solve like the IT factor module for QBs I described above.
Are there pain points that you were referring to that I didn't address in these roles you are describing? I am genuinely always interested to hear people's approaches to these things what we've built was pretty much purely out of doing many of these projects and automating our pain points
I'm not lauding him for his analytics work. Again - I came onto the initial analytics threads to simply show that he implemented a department in Carolina that didn't exist prior to refute the idea he shuns analytics and sees no value in them.
My visibility to NFL teams is mainly contained to the training room. So I'm only passing along that the Panthers and Giants have had metrics that they measure in the training room. And actually, the Giants were ahead of the Panthers there. The GM before Gettleman, Marty Hurney, didn't have any analytics in place. Gettleman brought that aspect to Carolina. The Giants had been doing it for years.
But that's just one small area of data that's being collected. Again - just another instance of data being gathered, which is why I said the larger debate is what is being done with the data.
Here is a huge one that goes into your luddite pattern of mis-characterizing things. How many times when I suggested he needed to be more invested in things like team construction and play analysis analytics did you say things like "you don't know what's really going on behind closed doors" and suggesting that his disciples that hire IT professionals to run analytics might also have programs that address those concerns. Things you insist on despite him literally making fun of the idea of those programs.
That's why I came to the conclusion that you must like smelling Gettleman farts. Because you continued to suggest that there very well might be advanced analytics going on despite very little evidence to support you point.
You back off that now to save face but the fact remains you charged into this debate without any kind of tangible knowledge of applied advanced analytics and argued with people that do.
Quote:
Nutrition and workout data is basically the absolute minimum required to check the box of "has an analytics department." I have no idea how anyone can use that to laud Gettleman or to suggest that he's actually embracing analytics. Especially when he very publicly mocked people using a data-based approach to evaluating talent and roster construction.
I'm not lauding him for his analytics work. Again - I came onto the initial analytics threads to simply show that he implemented a department in Carolina that didn't exist prior to refute the idea he shuns analytics and sees no value in them.
My visibility to NFL teams is mainly contained to the training room. So I'm only passing along that the Panthers and Giants have had metrics that they measure in the training room. And actually, the Giants were ahead of the Panthers there. The GM before Gettleman, Marty Hurney, didn't have any analytics in place. Gettleman brought that aspect to Carolina. The Giants had been doing it for years.
But that's just one small area of data that's being collected. Again - just another instance of data being gathered, which is why I said the larger debate is what is being done with the data.
Ultimately, I think both sides are right here, or at least there is room for both sides to be right - Gettleman (and the Giants, by extension) can simultaneously be open to some analytics (training/fitness/nutrition) and not pay enough attention (or flat out devalue) other analytics. It doesn't have to be binary, that someone is either a data junkie or a luddite, and the characterization as such definitely derailed this debate even when good points were being made.
The results seem to suggest that there is plenty of room for improvement, and some of Gettleman's personnel moves do appear at times to be grounded in the old school eyeball test more so than data. I think, as fans, we should hope that the front office continues to invest in the analytics side of the business. It doesn't have to replace the previous methodology; it can supplement it. More information is (almost) always a good thing.
You back off that now to save face but the fact remains you charged into this debate without any kind of tangible knowledge of applied advanced analytics and argued with people that do.
I've not backed off anything, because I've never asserted ANYTHING about advanced analytics. For the umpteenth time - this is what I've posted:
That's it. I made no claims about the extent of the analytics. Simply that they existed! An actual viewpoint that you said wasn't true!. There's backing off here - but it is you doing it, all while lobbying moronic insults.
Gatorade summed it up perfectly, and my stance actually agrees with him - that having analytics is a good thing and I hope that the Giants are doing their best in that regard. I've never made any claims about the Giants expertise in analytics or anything about advanced analytics, so spare me the bullshit. But then again - I haven't mined peoples LinkedIn profiles.
Your rants against Luddites are hysterical because tehy derive from one main post - that Gettleman implemented an analytic department in Carolina. Period.
This is the problem, you talk about Gettleman as if he understands game theory and you use that quote as support yet as many other than me have pointed out we are not even seeing simple game theory applied on the field.
And this is the larger point about you not understanding analytics. The points you so proudly attempt to make lack real basis when you are able to see it with the perspective of someone that really understands advanced analytics.
You are so over the top with this Game Theory talk that its laughable. But carry on, your posts are definitely entertaining.
But i'm glad you find how far we've fallen behind entertaining. I don't.
Eagles and Advanced Analytics New York Times - ( New Window )
As for the Giants, I have no desire to stand on a soap box and scream about game theory for days. You seem to enjoy it though, so keep up the good work!
As for the Giants, I have no desire to stand on a soap box and scream about game theory for days. You seem to enjoy it though, so keep up the good work!
It's funny that your lack of understanding the larger points that I am trying to make means I can't read or if you actually read the article I shared and comprehended it (again funny how you talk about my reading skills) you would be equally concerned.
EA has actually come back to have a substantive debate about this so he's no longer a luddite in my book. Maybe you can take his place!
I stopped reading your long diatribe posts a while ago. It just gets tiresome to keep reading the same things over and over. What exactly are you trying to prove and if you actually do turn me (or anyone else) to see the light, what will then happen?
I don't talk about Gettleman understanding anything. Those aren't even my words - they are from an article detailing what each team was doing for analytics. THEY are saying he and Rivera implemented game theory in their approach to 4th downs.
Again - for those the dense who can't parse what my message is:
I use that quote for one specific reason only - to show that Gettleman implemented an analytics department in Carolina.
Why have I had to re-emphasize this over and over again, and why are you failing to comprehend this very basic point?
My only real goal is at least here that is an accepted fact that we don't have the right people in place to be technology leaders we need to be a successful franchise the next 10-15 years.
Let's look at this as a game theory problem.
If Mara interacts with 1,000 fans a year and 100 of them are concerned about the direction of advanced analytics or at least mention it there is some probability that he does something about it.
Now let's say if this number grew to 200 or 300. This certainly increases the probability that he does something maybe by 2X or 3X, this might even be an exponential function based on what might be viewed as the acceleration of these requests or even as they say approach a critical mass of 50%.
Hell if these threads even convince 1 person that writes him to mention this I would have considered this a success from a game theory perspective.
This is where the fart smelling comes from. The lengths you go to be on these threads and to protect the supposition that Gettleman might be the wrong leader as it relates to the long term future of the team is astounding. It really is like a teachers pet that wants to constantly kiss the ass of a teacher, he's already in charge and you aren't really bringing any new information to the table other than the fact that you like him. We know that.
NoGainDayne : 11:09 am : link : reply
if you don't see how putting forth a quote over and over again that other people have presented evidence against being accurate as it relates to what is going on inside the Giants building is annoying
It wasn't even a quote about the Giants. It was about Carolina!! I never made any claims about what was going on with the Giants.
You really fucking suck at reading comprehension.
Ponderous.