How many solves in an average does it take to be Sub-X?

[Nir1213]

I wasn't meaning for me necessarily. I hear too many people saying stuff like "I'm sub-15" after one lucky solve, and I was wondering what people thought is the norm. I personally use multiple averages.

it goes for everyone else who says sub x (including me lol)

 

[HippieCuber]

Well if you are supposed to use ao1000, than some beginner cubes will not know what they average for a long time because at that speed you rarely have sessions over 12 solves

 

[joshsailscga]

Speaking qualitatively because I am no math person: I've always figured sub-x refers to, what am I most likely to average at any given time? Obviously the shorter the average the greater the possibility of variability. To give an example, I currently consider myself sub-12 because if I sit down to run an ao100, it will almost always be somewhere between 11.6-11.9. Now if I don't warm up or have an off day, the beginning of the set may average higher, maybe I'm recording mid-12 ao12s early on or might even see a low-13 ao5. Does that mean for that short period of time I am not sub-12? I would disagree. I know by the end of the 100 solves I'll have pulled back down below 12.
What sub-x does NOT mean imo is "I got a sub-12 ao5 so I'm sub-12 but most of my averages center around 14 seconds".

Obviously the slower you are, the larger the time differential. I'm currently learning Roux and while I have a sub-15 ao5 and a sub-19 ao100 in that method, I would not say that I am globally sub-20 yet because any given ao100 is likely to be a little over 20 seconds average.

 

[RiceMan_]

I'd say ao-100

 

[DNF_Cuber]

couple of millisecond

yep, if 470 milliseconds is a couple

 

[Eamon]

i say 500 is reasonable, you won't improve too much through 500 solves, and it's also a number that is representitive

 

[DNF_Cuber]

i say 500 is reasonable, you won't improve too much through 500 solves, and it's also a number that is representitive

that is what I go by for 3x3

 

[Nir1213]

Well if you are supposed to use ao1000, than some beginner cubes will not know what they average for a long time because at that speed you rarely have sessions over 12 solves

true, but for that i think ao12 is enough for them. When they get faster they can do like ao100 or 1000 solves.

 

[Owen Morrison]

No it isn't, because if you're improvement is very fast, then the average will include times when you were a lot slower. So in reality your true average will be faster the what the ao1000 says.

If you are sub 10 you won't improve that much in one ao1000.

 

[Nir1213]

If you are sub 10 you won't improve that much in one ao1000.

maybe, but what if you were implementing a different alg which is much better than yours, and the ao1000 will help reinforce that, thus getting your time down a bit. But i assume we are not talking about practice solves, and just raw solves, and in that case Owen is right.

 

[Owen Morrison]

maybe, but what if you were implementing a different alg which is much better than yours, and the ao1000 will help reinforce that, thus getting your time down a bit. But i assume we are not talking about practice solves, and just raw solves, and in that case Owen is right.

If you learn ONE better alg it either won't change your averages at all or such a minimal amount that it won't matter.

 

[Nir1213]

If you learn ONE better alg it either won't change your averages at all or such a minimal amount that it won't matter.

i know, thats why i said "getting your time down a bit"

 

[DNF_Cuber]

i know, thats why i said "getting your time down a bit"

if I learn an N perm that makes me .1 second faster when I get it (1/72 times) then I will be .0013 faster on average.

 

[qwr]

ao100 because I'm lazy.
unless you make a lot of mistakes that result in horrendous times then it won't make a big difference

do
a
statistical
test

Don't want to do that? Then stop saying "sub-x". It's a term so devoid of meaning and yet so popular in the community.

You could try to estimate a statistic like "90% of my solves are better than x" and if you did enough averages the CLT would probably kick in but I don't think there's any real point to it because no one really cares if your results are strictly correct in the frequentist framework.

 

[Nir1213]

if I learn an N perm that makes me .1 second faster when I get it (1/72 times) then I will be .0013 faster on average.

yeah i guess. anyway lets stop talking about this now