|
View Poll Results: Which system is best ? | |||
Simple average of top X equivs | 13 | 32.50% | |
Weighted average of top X equivs | 27 | 67.50% | |
Voters: 40. You may not vote on this poll |
|
Thread Tools | Display Modes |
05-24-2021, 02:54 PM | #21 |
FFR's Resident Trashpanda
Join Date: Sep 2012
Location: Michigan
Age: 29
Posts: 1,095
|
Re: Poll: Which global skill rating system is best ?
Issue there is the assumption that there will be too many files with that one specific skillset only with nothing else all also in the same small difficulty range, but way too many files would need to be added for this to be the case.
Last edited by Gradiant; 05-24-2021 at 02:55 PM.. |
05-24-2021, 03:53 PM | #22 | |
Forever Derbyless
Join Date: Aug 2020
Age: 32
Posts: 240
|
Re: Poll: Which global skill rating system is best ?
Quote:
This solution regardless how the engine and how FFR ages over time will be built to avoid these scalability concerns that will inevitably be present. This is why I'm advocating for the weighted solution. In the unweighted case, you have to have sufficiently enough people to play sufficiently enough files to be considered ranked. Is that guaranteed to happen? Who knows? Personally, I wouldn't hedge my bets on it if I know a solution that can circumvent these potential issues.
__________________
|
|
05-24-2021, 07:09 PM | #23 |
✘ Forever OP✘
Join Date: Dec 2008
Location: Canada,Quebec
Age: 29
Posts: 4,171
|
Re: Poll: Which global skill rating system is best ?
@People who voted for unweigthed:
I personally conceded the debate to WirryWoo after some more discussion on discord today. I invite you to take a look at the conversation that has been going on today (may 24th) in the #dev-chitchat channel. The demand I make after accepting the weighted system as a generator of our main skill rating metric is that we display, for each player, information that informs others of how flukey/biased their scores are. This (imo) would let the community easily find outliers which may not have been adequately ranked by the model (due to the high variance of opinion when tuning the hyperparameters), and better discuss the finegrained placement for events for example. Feel free to discuss things more on here or over on discord |
Currently Active Users Viewing This Thread: 1 (0 members and 1 guests) | |
|
|