Welcome to the Forum Archive!

Years of conversation fill a ton of digital pages, and we've kept all of it accessible to browse or copy over. Whether you're looking for reveal articles for older champions, or the first time that Rammus rolled into an "OK" thread, or anything in between, you can find it here. When you're finished, check out the boards to join in the latest League of Legends discussions.

GO TO BOARDS


@Riot: 10% Crit chance is actually 8%

Comment below rating threshold, click here to show it.

Ensign

Senior Member

03-08-2013

The initial set of 1922 data points is really, really weird.

When I'm analyzing a time series of binary data I like to look at how that data is distributed in time, to see how randomly the data points are distributed over the set. An example of how to do this is starting with the first crit, count how many attacks it takes before you see another crit. If the algorithm was perfectly random, you'd expect to see consecutive crits about 10% of the time, a single normal attack between crits 9% of the time, etc. In the initial sample we have 157 crits, so 156 crit pairs, which we'd expect to have a distribution of:

attacks before next crit: number of observations
1: 15.6
2: 14.0
3: 12.6
4: 11.4
5: 10.2
etc

What we actually saw:

attacks before next crit: number of observations
1: 1
2: 37
3: 3
4: 1
5: 1

Needless to say, that is an incredibly odd result. The entire set is like that, and there's no smooth function for generating random results where that isn't an incredibly odd result. Something is clearly wrong with the way Riot is generating crit results.


Comment below rating threshold, click here to show it.

Niqhtmare x

Senior Member

03-08-2013

nice, youre a junior wrenchman now?


Comment below rating threshold, click here to show it.

March of Dimes

Senior Member

03-08-2013

Quote:
Ensign:
The initial set of 1922 data points is really, really weird.

When I'm analyzing a time series of binary data I like to look at how that data is distributed in time, to see how randomly the data points are distributed over the set. An example of how to do this is starting with the first crit, count how many attacks it takes before you see another crit. If the algorithm was perfectly random, you'd expect to see consecutive crits about 10% of the time, a single normal attack between crits 9% of the time, etc. In the initial sample we have 157 crits, so 156 crit pairs, which we'd expect to have a distribution of:

attacks before next crit: number of observations
1: 15.6
2: 14.0
3: 12.6
4: 11.4
5: 10.2
etc

What we actually saw:

attacks before next crit: number of observations
1: 1
2: 37
3: 3
4: 1
5: 1

Needless to say, that is an incredibly odd result. The entire set is like that, and there's no smooth function for generating random results where that isn't an incredibly odd result. Something is clearly wrong with the way Riot is generating crit results.


Well, it's not random, it's pseudorandom, in which your crit chance increases slightly upon missing crits & decreases on hitting crits, in an attempt to smooth out the randomness. Perhaps you retain an overcharged crit rate for a period of a few autoattacks, explaining the 1s and 3s in your results. What would such an analysis look like if you disregarded those data points? Is 37 fairly common as a non-1 number?


Comment below rating threshold, click here to show it.

Bloodpaly

Senior Member

03-08-2013

Quote:
Bland:
Don't you think that's a little bit assuming?
Also "inaccuracy" is one thing... but looking at this, it seems to me there's something rather blatantly wrong with the code; and not necessarily something with their RNG-system. I mean, sure, RNGs aren't perfectly random, but they're at least good enough not to be off by almost 20% of the expected value.

I just don't get how Riot could input a psuedo random crit system without doing any tests or simulations on its accuracy.

As far as I can tell, either Riot knew it was inaccurate when implementing the system and didn't care, or didn't test it all all, or something happened to the algorithm afterwords.


Comment below rating threshold, click here to show it.

Bland

Senior Member

03-09-2013

Quote:
Bloodpaly:
I just don't get how Riot could input a psuedo random crit system without doing any tests or simulations on its accuracy.

As far as I can tell, either Riot knew it was inaccurate when implementing the system and didn't care, or didn't test it all all, or something happened to the algorithm afterwords.


As long as we don't know 100% what's causing it, it could always be some extrordinarily small detail that may not have been tested by Riot (after all a lot of things in these tests are static... as a (pretty stupid) example: What if that bug only applies when hitting Mundo?).
At this point, you can consider it a ****-up by Riot either way, since obviously they've done a mistake, but that's no reason to just jump to conclusions that imply intent, when it could still as easily have been overlooked.


Comment below rating threshold, click here to show it.

Bland

Senior Member

03-09-2013

bump


Comment below rating threshold, click here to show it.

Bland

Senior Member

03-09-2013

another little bump


Comment below rating threshold, click here to show it.

I play blind

Senior Member

03-10-2013

no roit response? roit pls


Comment below rating threshold, click here to show it.

IckerBicker

Senior Member

03-12-2013

bump


Comment below rating threshold, click here to show it.

Bland

Senior Member

03-12-2013

bump