I’ve written about hockey analytics for 15 years. In that time, I always wanted a site that dug into not only IF a player is good, but when, why, and how. Something to help writers and fans tell those stories.
3 months ago I started building it
... and today, it's live!
@DavidCastilloAC I appreciate the feedback. Again I don't agree with that description because it explicitly says it's AI generated so I'm not sure how that shows as pretending, but I understand that you're basically saying bad writing is worse than no writing.
Also I've been churning stuff lately for the website and now this, so I think it's important to be transparent in case any new followers don't know: AI played a BIG role in making this. I wrote almost no actual code.
Basically I describe the whole thing: what features should it use, what literature should it look for (sometimes giving specific examples from memory), what techniques should it consider and how to test them, and what is ultimately the desired output.
Beyond that, I have an AI infrastructure than has "rules" that the agents must always follow including ensuring nothing is adopted without 1) proving it works empirically, 2) running it by me.
I had the image below in my head from the beginning as well at how to make it. But 2 years ago it would've taken several weeks to hard code all of it and test everything. I don't have that time. But I do have time to give very specific instructions to someone who's an excellent programmer and an okay thinker to take it over for large chunks of time until it needs my input.
IMO this is what AI is good for. Taking away the clerical grunt work and leaving the interesting thought provoking problems to the human. Some may disagree. And that's legitimate as well.
@DavidCastilloAC Right and I'm adding you what it would be like in those other 2 scenarios and which is the best option.
I don't have time to do everything. So the options are, don't publish until everything is ready, publish stats with no writeup, or punish stats with AI writeup.
Big Data Cup is up! https://t.co/TLlPBLZ4x6
Computer vision data of goalie AND puck. (See video) Please send to anyone that wants to work with data + students hoping to learn. Sports Performance Lab @ MLSE, @HockeyAnalytics & @UofT
@MeghanChayka@0xKeef Absolutely! But next time would it kill ya to get me some stick coordinates!? I'm making some really wild assumptions about how the human hand works to get this five-hole covered 😅
@0xKeef@MeghanChayka I disagree. I mean sure there's a bunch more you can do with more data, but I think that misses the point of Big Data Cups where are more like proofs of concept. There's plenty of information here to be able create visuals that accompany theories on things that might matter.
Concurred. When I was getting @PostHockeyHQ up and running last month, @AlisonL was one of the first people I reached out to because I knew that if my goal was to get people talking about analytics in a new way, she'd be the one of the best people to talk to.
@DavyJonesLR@AlisonL Alison has done an amazing job with education to the general public on hockey analytics and her Analytics with Alison series is great
Btw it takes a while to get it to look like that. If you pop open the hood there are some hideously accursed images like this that are needed to inform the way that the thing moves and, more importantly, how it /could/ have moved.
@TheNJDAdvocates My version of these takes are a slideshow that's been in my head for years and I never had the energy for the encyclopedic breakdown. Bravo on having the patience lol.
It's so much fun to just click around the Goals page and watch some of these plays develop (speed in the neutral zone, empty net goals, set plays, OT winners, etc).
@DavidCastilloAC Out of curiosity, would you rather a) Not have the feature, b) Have the feature, but not have a writeup, or c) Have the feature, and the AI writeup that you see on the site right now?
Thanks, I appreciate the support. And yeah I'm not under the impression that the writing in the methodology section is good and it explicitly says that it was written by AI as a placeholder.
I disagree that it isn't transparent, though because the average stats website on the internet keeps the majority of their work completely opaque and even those with writeups don't shot you how they get there, some don't show their validation, some aren't clear if they retrained on the full dataset or not, etc.
These placeholders may not be a good read, but the fact that I haven't edited it should give you even more confidence that what's there is exactly what you see on the site. By the time the site exits Beta (another thing that is transparent and visible on the top banner no matter where you are in the website), those writeups will be human-reviewed. In the interim, there's only one me and I'm sure you can understand that building the website's infrastructure is a higher priority than making the writeups page-turners.