Old 08-23-2019, 01:47 PM   #1
The Dominator
D7 Elite KeysmasherFFR Veteran
Dynam0's Avatar
Join Date: Sep 2005
Location: North Bay, ON
Age: 30
Posts: 8,679
Default Learning Rates in Games

I guess it's been a while, how's everybody doing lol

So I got really bored at work and played Kitten Cannon for a whole day, recording each score and trying to determine what a reasonable theoretical limit would be. After 1000 plays I ended up with a high of 2082 and this seemed like a good amount of data to analyze. Unfortunately I quickly found myself lacking in the statistics department as I started off using a normal distribution model rather than an exponential decay model.

Anyways I soon gave up because I realized I just spent an entire day at work doing nothing but analyzing Kitten Cannon score but found a neat paper in the process:

Essentially this study looks like it was designed to inform game developers on how best to manage high scoring models depending on how they want their players to feel rewarded etc. It's interesting because it takes into account the learning aspect in games and how that plays a role in obtaining a higher score versus no learning curve.

It would be interesting to do a similar study with FFR scores (obviously not me but someone with a statistical background) and see where this game and its players stand on the learning rate spectrum. I think we all agree that relative skill increase exponentially decays over time and this is quite dramatic.
Dynam0 is offline   Reply With Quote
Old 08-23-2019, 03:07 PM   #2
FFR Veteran
Dinglesberry's Avatar
Join Date: Dec 2007
Location: Ontario, Canada
Posts: 2,549
Default Re: Learning Rates in Games

I think it's a good concept in theory but I don't think you can really apply it too much to a rhythm game since a game like this requires warming up, is physically demanding, you get tired etc, but maybe it would still fit... Getting high scores here requires a combination of you feeling well physically and mentally compared to games like those where the physical demand isn't very high.

With enough plays though maybe it would still fit a sqrt(x) graph or whatnot.. I mean, based on how skill progression works I would assume that your rate of improvement in terms of skill in a game is sqrt(x) (where x is number of plays) whereas the actual difficulty of improving is exponential x^2 (with x being time). For most games, improvement at a higher level is more difficult because mechanics of most games at that point just end up being micro-optimizations etc, whereas a game like FFR doesn't necessarily end up being more complex mechanically, it just requires higher focus/speed.

I think a better metric would be measuring the amount of time taken to reach x level, rather than number of games played.

Last edited by Dinglesberry; 08-23-2019 at 03:08 PM..
Dinglesberry is offline   Reply With Quote
Old 08-23-2019, 03:20 PM   #3
~ お ま ん こ ~
DaBackpack's Avatar
Join Date: Mar 2014
Location: unlikely
Age: 26
Posts: 882
Send a message via Skype™ to DaBackpack
Default Re: Learning Rates in Games

Yeah, you'd be surprised that reward mechanics in video games is a serious academic discipline. (Often in terms of serious games, or crowdsourcing games where intrinsic reward might be minimal)

It's not quite the same as the paper described, but one of my lab mates is writing her dissertation on a systematic exploration of the kinds of reward signals that players are most receptive to for... grindy, perhaps monotonous games where intrinsic reward is minimal. That is, games that might not be "fun" in a traditional sense.

Her ontology is "narrative reward", "leaderboard/score reward", and "collection/customization reward", just going from memory. Part of her research was about balancing and scaling score-based reward signals to encourage replay. I'll have to ask her what she eventually landed on. I'd imagine that a logistic curve would suit a general audience, since diminishing returns at the upper echelons of skill:

1) Suggest an upper bound for maximum performance, something that you make incremental progress towards. Some games have HUGE score differentials between skill tiers, which makes it impossible to determine how close your given score is to an optimal "perfect score"... or more importantly, how much skill you require to reach this.

2) Newer players are less likely to be discouraged when they see their scores in the middle of the logistic curve. How disheartening can it be to see that your starting score is orders of magnitude below some higher scores? With a logistic curve, your score of 0.5 is still in the ballpark of 0.85, so you are more likely to try to improve your score to reach this landmark because it is actually "in reach". (Of course, the skill differential doesn't change, it's just a matter of how numbers can be used to encourage or discourage mindsets of personal improvement).

But I don't play FFR so I don't know what the common perspectives on scoring curves are around here

Originally Posted by TWG Dan Hedgehog View Post
there are 743 matches for hedgehog suicide on deviantart
that's kind of a sad statistic
DaBackpack is offline   Reply With Quote
Old 08-23-2019, 06:21 PM   #4
Carry your failures proud
FFR Veteran
Soundwave-'s Avatar
Join Date: Sep 2015
Age: 19
Posts: 634
Default Re: Learning Rates in Games

There's a correlated factor in here, which is that ones score on a particular file diminishes increasingly faster with a linear decrease in skill.
Originally Posted by [11:38 PM] Hakulyte
only person who can legit tilt me is like YoshL
Originally Posted by スンファンさん
Soundwave- is offline   Reply With Quote

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump

All times are GMT -5. The time now is 03:54 AM.

Powered by vBulletin® Version 3.8.1
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.
Copyright FlashFlashRevolution