Further statistics of Catagolue based on textcensus.csv

For general discussion about Conway's Game of Life.
Post Reply
Dylan Chen
Posts: 114
Joined: March 27th, 2020, 8:07 am
Contact:

Further statistics of Catagolue based on textcensus.csv

Post by Dylan Chen » March 5th, 2021, 4:48 am

Catagolue' textcensus page is very useful, it listed out every objects we have and the total occurrences.
textcensus/b3s23/asymmetric-soups

when exploring the table of xs20 synthesis, I found that all remaining 12 xs20 have no natural soups. But much to my surprise, among 112k possible xs20s, only 21k of them have occured naturally. That means, even for the small still-life like xs20s, which we are about to synthesis for all, 81% of them haven't natural occured.

For the xs20s we have, the distribution is extremly uneven too.
Image
the most frequent xs20 occuered 553 million times, more than every xs20 else combined.

Image
6000+ of xs20s only occur 1 time. 5.3k(4.5% in total) have the occurrences greater than 20. If you chose a random xs20 from all 112k possible candidate, 95% of chance it would be counted as your 'discovery'.

Code: Select all

(sry for the 'spamming', because of the bad network. I have to submit in case the connection failure)
now edited as yujh suggests

You can see the xs19s have almost the same kind of distribute. 
[img]https://cdn.discordapp.com/attachments/404518331605975040/816857562024378398/unknown.png[/img]

for the xs25s, things goes more extrem
[img]https://cdn.discordapp.com/attachments/404518331605975040/816856888109039676/unknown.png[/img]
more than half of known xs25s only occured for once.

the xs15 iseems more evenly distributed .yes of course,nearly all xs15s soups has been found. [url=https://catagolue.hatsya.com/statistics]1343/1353[/url]
[img]https://cdn.discordapp.com/attachments/404518331605975040/816856233196912640/unknown.png[/img]


also we have the plot of xp2 and xp3:
[img]https://cdn.discordapp.com/attachments/404518331605975040/816862288073981972/unknown.png[/img]
[img]https://cdn.discordapp.com/attachments/404518331605975040/816862461449076758/unknown.png[/img]


So I did some other digs, towards other patterns.

we have:
4254 different xp2,
889 different xp3,
96 different xp4.
yes, it is. There only have 96 types of xp4 in all 2.38*10^14 asymmetric soups.

the total occurrence:
xp2, 3.17*10^14
xp3, 76125552460
xp4, 99103161
In asymmetric soups, xp2 is 4000 times more frequent than xp3.
xp4 show up one in a thousandth to xp3.


For all asy-soups, we have 1094967510989987 objects, in 241898 different types.
But the top 10 common objects, consisted of 99.1% total objects.
the block and blinker combine, show up 621747462979145 times, that is the 60% of all objects.

you can check the statistics page of Catagolue, exploring the 'Common objects' table


from the discord img,
if we list the most common object in each type/class, like the glider for xq4, the pulsar for xp3.

we will find that:
the most common xp15(Pentadecathlon) is more frequent than the most common xs15 (Moose antlers)
glider produce switch engine(yl384) and the block-laying switch engine(yl144) are more common than any xs20, xs13, xs19.
and you cannot imagine the most common xs13(Beehive at loaf) is rare than xs14-18 and xs20.

both xp30 (trans-queen-bee-shuttle) and (cis-queen-bee-shuttle) are more common than any xp5. the most common xp5(Octagon 2) only have the half occurrence of trans or cis-queen-bee-shuttle.

the number of methuselah_25k is greater than any xs25. the megasized_30h is frequent than any xp30.



in this pic, we can inspect the latter part of table.

xq7 Loafer, which has occured for 5 times, is more 'frequent' than xp16 (Rob's p16 4 times) and xq12 (Schick engine 3times).


some statistical inference:
if rare objects obey exp distribution,based on current 5 Loafer, it will cost '910 days' to get next Loafer in C1.


Image
1haul = 10^7 soups
Image

you can check the the number of soup to get a certain type of object.(not a new discovery)






Mateon1_32x32_Test
early statistics of 32x32 soups verses C1 16x16 soups. (based on 14billion objects we currently have)
1. Top 30 frequent object didn't change their sequence . Some of the top 50 switched position.

2. we have less frequent block-laying switch engine/glider-producing switch engine in 32x32, comparing with certain number of total objects.
But for same number of soups, 32x32 has more possibility to produce those two switch engines. (180% yl144 to 16x16)

3. if we take one 32x32 soup equal to four C1 soups. the objects we have is 15% less than C1, especially for gliders(30%), XWSS(20%).
which suggests 32x32 is not effeciency as 16x16 C1 soup search in finding xq4 spaceships.

Conclusion: no special benefits observed from 32x32 random soup search. waiting more data to confirm that.





The occurrance of asy 16x16 Methuselahs are log-linear:
Image


by inaccurate statistical inference, there would be 8.6x10^67 methuselah_25k in future total 10^76 asy-soups.
Image
with in my unreliable statistical regression, the most longevity meth would be 288k.


My question is: Does this reveal the boundary of CGoL universe?
From 16x16 soups, you can go at most 288k ticks.

And in philosophy: We would exhaust our universe to finish all 16x16 soups, while the most we can get is just merely 288k ?
Last edited by Dylan Chen on March 6th, 2021, 1:51 am, edited 2 times in total.
Tools should not be the limit.
Whether your obstacle is a script, an stdin, or Linux environment computing resouces.
check New rules thread for help.

Dylan Chen
Posts: 114
Joined: March 27th, 2020, 8:07 am
Contact:

Re: Further statistics of Catagolue based on textcensus.csv

Post by Dylan Chen » November 26th, 2021, 9:57 am

inspirehep — 03/29/2021
for 32x32 soup search

early statistics of 32x32 soups verses C1 16x16 soups. (based on 14billion objects we currently have)
1. Top 30 frequent object didn't change their sequence . Some of the top 50 switched position.
Image

2. we have less frequent block-laying switch engine/glider-producing switch engine in 32x32, comparing with certain number of total objects.
But for same number of soups, 32x32 has more possibility to produce those two switch engines. (180% yl144 to 16x16)

3. if we take one 32x32 soup equal to four C1 soups. the objects we have is 15% less than C1, especially for gliders(30%), XWSS(20%).
which suggests 32x32 is not effeciency as 16x16 C1 soup search in finding xq4 spaceships.

Conclusion: no special benefits observed from 32x32 random soup search. waiting more data to confirm that.
Tools should not be the limit.
Whether your obstacle is a script, an stdin, or Linux environment computing resouces.
check New rules thread for help.

User avatar
dvgrn
Moderator
Posts: 10612
Joined: May 17th, 2009, 11:00 pm
Location: Madison, WI
Contact:

Re: Further statistics of Catagolue based on textcensus.csv

Post by dvgrn » November 26th, 2021, 11:15 am

Dylan Chen wrote:
November 26th, 2021, 9:57 am
Conclusion: no special benefits observed from 32x32 random soup search. waiting more data to confirm that.
Interesting report! I'd expect the benefits, if any, to be very subtle. It's true that the perimeter-to-area ratio is lower for 32x32 than for 16x16, so you wouldn't expect to see an increased proportion of anything that needs to escape outward from an edge into empty space.

What I _would_ expect to see a lot less of is "index fossils" -- objects that are would ordinarily be very unlikely, but they have one particular low-population bottleneck pattern that generates them after some number of ticks (doesn't matter how many). When the same index fossil appears in multiple soups, it's a sign that the two soups have converged to the same bottleneck pattern.

16x16 is small enough that some significant percentage of soups that get tested end up duplicating the end of some other soup's evolution -- usually due to death by overpopulation in the first few ticks, resulting in just a few small active regions. Seems like at 32x32, the percentage of duplicates of this kind should be very much smaller. (?)

hkoenig
Posts: 258
Joined: June 20th, 2009, 11:40 am

Re: Further statistics of Catagolue based on textcensus.csv

Post by hkoenig » November 26th, 2021, 7:41 pm

Long time ago when I was doing random bitpattern studies, I noticed similar effects, didn't do much else to follow up. But this matches my experience, for what that's worth.

For a while, I tried some large areas, up to 1024 square. One thing I noticed was that it seemed like a lot of the Ships came from B-heptomino/Herschels that formed along the edges and survived because of that isolation. Just about every run had one or more of that familiar Block/Ship pattern along the edges.

It might be interesting to compare different densities for the 32x32 case, (or larger areas) and see if it makes a difference. Another idea I had as a way to try and remove some of these boundary effects, but never pursued, was to try a variable density/probability based on the distance from the central core of the array. Thin out the array instead of having an abrupt edge.

User avatar
dvgrn
Moderator
Posts: 10612
Joined: May 17th, 2009, 11:00 pm
Location: Madison, WI
Contact:

Re: Further statistics of Catagolue based on textcensus.csv

Post by dvgrn » November 27th, 2021, 8:52 am

hkoenig wrote:
November 26th, 2021, 7:41 pm
It might be interesting to compare different densities for the 32x32 case, (or larger areas) and see if it makes a difference. Another idea I had as a way to try and remove some of these boundary effects, but never pursued, was to try a variable density/probability based on the distance from the central core of the array. Thin out the array instead of having an abrupt edge.
I like the first idea especially -- there's probably a starting density that doesn't cause such a big drop at T=1 due to overpopulation, which would (I'm guessing) produce a few less duplications due to early low population bottlenecks.

However, it's not like I've seen a large number of "index fossils" relative to the total number of soups that are being processed. Maybe these soups that converge to duplicates are such a vanishingly small percentage of the total, that it doesn't really matter if they get adjusted to be an even more vanishingly small percentage.

hotdogPi
Posts: 1587
Joined: August 12th, 2020, 8:22 pm

Re: Further statistics of Catagolue based on textcensus.csv

Post by hotdogPi » November 27th, 2021, 9:22 am

dvgrn wrote:
November 27th, 2021, 8:52 am
However, it's not like I've seen a large number of "index fossils" relative to the total number of soups that are being processed. Maybe these soups that converge to duplicates are such a vanishingly small percentage of the total, that it doesn't really matter if they get adjusted to be an even more vanishingly small percentage.
There are currently 32,775 methuselah29ks; exponential distribution predicts about 26,000. Lidka is definitely messing with that.
User:HotdogPi/My discoveries

Periods discovered: 5-16,⑱,⑳G,㉑G,㉒㉔㉕,㉗-㉛,㉜SG,㉞㉟㊱㊳㊵㊷㊹㊺㊽㊿,54G,55G,56,57G,60,62-66,68,70,73,74S,75,76S,80,84,88,90,96
100,02S,06,08,10,12,14G,16,17G,20,26G,28,38,47,48,54,56,72,74,80,92,96S
217,486,576

S: SKOP
G: gun

User avatar
dvgrn
Moderator
Posts: 10612
Joined: May 17th, 2009, 11:00 pm
Location: Madison, WI
Contact:

Re: Further statistics of Catagolue based on textcensus.csv

Post by dvgrn » November 27th, 2021, 10:51 am

hotdogPi wrote:
November 27th, 2021, 9:22 am
dvgrn wrote:
November 27th, 2021, 8:52 am
However, it's not like I've seen a large number of "index fossils" relative to the total number of soups that are being processed. Maybe these soups that converge to duplicates are such a vanishingly small percentage of the total, that it doesn't really matter if they get adjusted to be an even more vanishingly small percentage.
There are currently 32,775 methuselah29ks; exponential distribution predicts about 26,000. Lidka is definitely messing with that.
There was some sorting work done on that a few years ago -- see here and here. I didn't carry the analysis very far, but it definitely seemed to be true that Lidka was the cause of the big bump in the 29K bin, and there wasn't anything similar happening in the 30K bin.

At this point there's a lot more data available in the higher bins. I haven't looked to see if there are any more interesting bumps in the predicted values, that could be tracked down to find other common bottlenecks. But anyway, maybe the main point is that it does seem that this kind of bumpiness would happen a lot less with 32x32 soups than with the current 16x16 ones.

Post Reply