Discussion in 'Leaguegaming American Hockey League (LGAHL)' started by bigdiskk, Nov 10, 2019.
I'm first on my team for something!
Too many numbers to say I’m a shitter
This is awesome! These numbers actually mean something! Well done @bigdiskk
Yes, bad quality of teammates and play the hardest teams is a recipe for disaster lmao. RIP, nice job tho OP!! I'm a finance major so spreadsheets make me happy!
The thing is tho, if a player and his linemates happen to have only played top lines on top teams, and always lost by only 1 goal each time, then they wil look absolutely attrocious on the ELO charts. But they could also be the best second line in the league had they had slightly easier matchups.
Obviously this is an extreme case, but the bias is real. So context must be used in every case.
What I'm thinking of future iterations of my data-crunching that would do things differently:
Keep the ELO calculation as-is, but introduce a secondary CR calculation where I could weigh down ELO by score difference after 3 periods. For instance, Winning/Loosing by 5 should have a bigger CR impact than Winning/Loosing by 1. The issue I have with this tho is teams will usually go 'all out' at the end of the third to try and get that equalizer, and quite often a very tight game ends up with a 3 goals difference. So it could bias things even more? Might not be a good idea haha after all haha.
Calculate CORSI. In the real NHL CORSI is shot attemps (including blocks and missing the net), but my implementation would only include (shots + sum of opponent blocks). The resulting RELATIVE CORSI % values could be used as a factor to modulate the CR gains / loss.
Calculate TOA% and use it as a factor to modulate the CR gains / loss.
And finally, I'm not ever taking 32+ screenshots again. That shit was wayy too time consuming. I'll just generate a webpage with every table for every team (even add some charts), and save that to a monster PNG or PDF and share this instead.
@WingelsDGC I'd be interested to know what you think about that. I've went back and saw some of your work with ELO and that's fucking crazy. Maybe we could collab and make a weekly report for each leagues?
Excellent, yes I have been a few games where it was tied up until the last 10 minutes then they start getting EA bounces and momentum so then they run away with it. Very cool though, If you need assistance I would gladly help.
I vote that bigdisk and wingles collaborate to rewrite the lg power ranking since those seriously broken
come at me fuckers, ill fking shutdown you. I live for the biggies
Jeeeze my guy ^^^
Love the analysis my friend, you obviously have solid knowledge on coding and what not(which I don't have!).
Point #1 is excellent, as teams who win more dominantly should have an advantage over those who always have close games. Obviously QoC comes into play but that's why you have your smart calc's.
Corsi can't work in Virtual NHL, as it all depends on a teams playstyle. Some teams like to only shoot when they think it'll go in the net, while others like to get the puck on net more and create goals that way. There isn't a unified way to play hockey in NHL20, so it would create too much disparity in the numbers I think.
Point 3 is awesome, and something that I like to look for in players.
With this fantastic analysis, clearly I need to step up my game and put some more information into my posts. Maybe i'll do one next week
This is fucking sweet good stuff
^^ This is the collaboration we never knew we needed
Great work bro! @RyBish33 that is your fault! But it is ok we all love you!
Thanks for your feedback!
My initial approach was to use data that is unequivocal to generate unbiased performance indicators.
That’s why using only Wins or Losses as a single source of truth is so powerful.
So I think i’ve changed my mind (for now) and will keep everything as is; I don’t want to convolute the model and add any additional factors to compute the output. So i’ll put the idea to use shots or TOA or even goal differential to adjust the data on ice for now.
I also realized that since every year iteration is a different/semi-new game, (and that not every skill transfer), that not looking prior to the season in progress is actually a feature; not a caveit.
I’ll try to find some time tomorrow or friday to automate the work a little more and make it easier to output results weekly. Next week I’ll try to make better visualization to make the data easier to digest.
The story telling will still need some work so if anybody wants a to collaborate to make it even better I’m down
Really great work and thank you for putting the time in for this btw. It would be really cool to collab with Wingels with the ELO ratings that he uses for draft prospects such as SF and efficiency to not sway your work or change the diagnosis you have done but just as another form to paint a full and complete picture.
B R U H
Listen what ever puts me #1 in something is good in my book
Scrub hunters exposed!
Nice crunching bigdiskk, you are the analyst LG needs right now.
this is dope my man
Separate names with a comma.