The Online Life-Like CA Soup Search

[Lewis]

Are any new rules going to be added?
I'm going to request b35/s23 (I think I might have said already) because after doing some researching by hand, I found some interesting oscillators and 2 spaceships in it (a c/2 and a 2c/5).

Also, will tere be an option to show a table for oscillators only (like there is with spaceships at the moment)?

[Nathaniel]

Are any new rules going to be added?
I'm going to request b35/s23 (I think I might have said already) because after doing some researching by hand, I found some interesting oscillators and 2 spaceships in it (a c/2 and a 2c/5).

Also, will tere be an option to show a table for oscillators only (like there is with spaceships at the moment)?

Yep, I'll add plenty of new rules after I get all the features I want up and running. It's just a time thing. And yep -- you can now choose any combination of oscillators, still lifes, spaceships/puffers, and include/exclude pseudo objects.

[Lewis]

I've noticed the 2x2 results go down to patterns with 50 or more occurences, but HighLife's go down to 10+ occurences. Is there any particular reason for this?

[Nathaniel]

I've noticed the 2x2 results go down to patterns with 50 or more occurences, but HighLife's go down to 10+ occurences. Is there any particular reason for this?

Because I'm still working on things

[Nathaniel]

OK, version 1.02 of the Soup Search is now out. The main new features have already been half-released online in bits and pieces over the last week, and they are:

• You can now view objects in the census by type (oscillator, spaceship, etc). Similarly, you can include or exclude pseudo-objects (they are now excluded by default).
• The download and frequently asked questions pages are now actually online. Not that they'll be particularly useful for the people who have been paying attention to this project. Let me know if you think of more questions that would be good for the FAQ though.
• I added B35/S23 to the list of supported rules.

The main updates to the script itself include of course some bugfixes, and a break-free condition so that (for example) if you find a ridiculously-long lived replicator-based pattern in HighLife, it won't get stuck there forever. Also, the census uploading routine is much more stable and quick now (it would sometimes hang for about a minute before or even stop uploading altogether -- now it maxes out at about 6 seconds for me). I've been running the script constantly for about 72 hours now and it seems to be pretty good now (hasn't gotten stuck or anything like that in that test time).

[iconmaster]

Will you eventually add support for Generations rules?

[Nathaniel]

Will you eventually add support for Generations rules?

I don't see why not, but von Neumann Rules will surely come first.

Also, apologies folks; apparently my internet right now is good enough to post on forums and corrupt files, but not good enough to upload new versions of files. Thus, the census and long-lived patterns results pages might be unavailable for a little while.

[iconmaster]

I would like to see Diomeba and Balanced Ameoba in here, as well as 238/3.

[Macbi]

We have a new leader for the life methuselahs, with a lifespan of 21813!

[Nathaniel]

We have a new winner again! Lifespan 24389.

[Macbi]

I didn't think you'd stay beaten for long. I was rather surprised at the time, Veronika had searched a tiny fraction of the number of soups you had gone through.

[Lewis]

Would it be possible to make it so that the number of objects displayed can be changed, similar to the object types which can be changed now?
It could maybe have settings for showing all patterns, or just ones that have occured a certain number of times.

[Nathaniel]

Would it be possible to make it so that the number of objects displayed can be changed, similar to the object types which can be changed now?
It could maybe have settings for showing all patterns, or just ones that have occured a certain number of times.

If what you have in mind is making it so you can see patterns with fewer occurrences than is currently the cutoff, then unfortunately the answer is no. The cutoff, when it gets high for some rules (like it is for 2x2 and move right now, I'll try to rectify that today) it's due to me being behind on my cleaning of the data (I have to tell the database which objects are pseudo objects and even split some objects up into two different objects).

[Lewis]

Would it be possible to allow me or other forum members to help in fixing the data?

[calcyman]

Can you search for even-symmetric random soups, in order to find weekender predecessors?

[dvgrn]

Can you search for even-symmetric random soups, in order to find weekender predecessors?

Interesting idea... I spent some time thinking about this yesterday, but the more I thought about it the less I could see the point. You'd expect a weekender to appear along the axis of symmetry, I take it? Only two possible output directions per pattern tested. Even with the even-symmetric criterion, I suspect the sun would grow cold before a weekender would happen to crawl out of a 20x20 soup.

-- Would the census script recognize that something exciting had happened, by the way, or would a weekender be just another settled-down predictable pattern?

Even if a weekender does appear, the odds are pretty lousy that a new construction recipe could be worked out based on the initial soup -- 20x20 random fields aren't particularly synthesizable as a rule. You'd have to get extra lucky and go through a glider-constructible stage on the way to the weekender. Otherwise, might as well just use one of the lifesrc variants and cook up as many weekender "soups" as you want from scratch:

Code: Select all

#C 7-tick predecessor of a weekender fits in 20x19
x = 20, y = 19, rule = B3/S23
3b2o2bo4bo2b2o$2o4bo6bo4b2o$5b2obo2bob2o$3o2b2obo2bob2o2b3o$5bo3b2o3bo
$2o6b4o6b2o$bobo2bobo2bobo2bobo$bo4bobo2bobo4bo$2ob2ob3o2b3ob2ob2o$2o
5b2o2b2o5b2o$3o2bo8bo2b3o$2o2b4ob2ob4o2b2o$b2ob2o8b2ob2o$b2o6b2o6b2o$
3bobo2b4o2bobo$b2o4bo4bo4b2o$b2o2bobob2obobo2b2o$2b4o8b4o$bob2obo6bob
2obo!

... Hmm, suppose instead of starting with a random field, the census script chopped out a random 20x20 (or whatever) chunk of burned-out ash from the previous experiment, added one or more incoming gliders or *WSSs, and ran a new experiment from there? Might get a higher proportion of fizzle reactions -- not even sure about that, though, come to think of it -- but when interesting stuff cropped up it'd be pretty much guaranteed glider-constructible.

Might make sense to increase the initial size a bit, since there aren't nearly as many still-life configurations as there are 20x20 soups. But then again, I'd say a 2^400-sized search space is considerably more than adequate... probably wouldn't run out of ash-squares to test even if the entire Slashdot readership starts running Golly soup searches.

[knightlife]

Can you search for even-symmetric random soups, in order to find weekender predecessors?

How about a symmetric glider-only soup search, a glider synthesis soup if you will. Purists will not like it though because of all the trapped gliders in the soup. How about four "random salvos" all aimed at a central 20 x 20 spot? Anyoue try this random symmetric glider synthesis yet? The "glue" program is a glider synthesis search program but only with slow salvos and only from one direction. The size of the salvos and the spacing between gliders could be selectable for experimentation (e.g. "optimized" for weekender). Census could be taken "as you go" for long-lasting salvos, but the time may be better spent searching the next initial pattern. Weekender may never appear but something very cool might! I would like to try "quad symmetric" salvos where the four salvos are simply 90 degree rotations of each other.

[Nathaniel]

New recordholder with lifespan 25177!

Also, I indeed plan to make other "modified" searches in the same vein as this. I just have lots of things that I'm working on.

[knightlife]

This works! This P6 oscillator came up pretty quickly, even manually. The trick seems to be to keep up the heat in the center.

Quad salvo synthesis of P6:

Code: Select all

x = 102, y = 102, rule = B3/S23
bobo$2b2o97bo$2bo96b2o$100b2o13$20bo$21b2o$20b2o2$83bobo$83b2o$84bo4$
30bo$28bobo$29b2o43bo$73bo$33bo39b3o$34bo$32b3o34bo$69bobo$69b2o3$58bo
$58bobo$58b2o2$38bo$39bo$37b3o$45bo$46b2o7bobo$45b2o8b2o$56bo7$45bo$
45b2o8b2o$44bobo7b2o$56bo$62b3o$62bo$63bo2$42b2o$41bobo$43bo3$31b2o$
30bobo$32bo34b3o$67bo$26b3o39bo$28bo$27bo43b2o$71bobo$71bo4$17bo$17b2o
$16bobo2$80b2o$79b2o$81bo13$2o$b2o96bo$o97b2o$98bobo!

The P6 oscillator is not new, but the synthesis might be.
I like this, no need to painstakingly search for the glider synthesis of the result!

[dvgrn]

This works! This P6 oscillator came up pretty quickly, even manually. The trick seems to be to keep up the heat in the center.

Hmm, not all of those gliders are doing anything useful, though:

Code: Select all

#C 16-glider synthesis of p6 oscillator, reduced from 24 gliders
x = 102, y = 102, rule = B3/S23
bobo$2b2o97bo$2bo96b2o$100b2o13$20bo$21b2o$20b2o2$83bobo$83b2o$84bo8$
33bo$34bo$32b3o34bo$69bobo$69b2o2$44bo$42bobo$43b2o4$64bo$63bo$63b3o
13$36b3o$38bo$37bo4$57b2o$57bobo$57bo2$31b2o$30bobo$32bo34b3o$67bo$68b
o8$17bo$17b2o$16bobo2$80b2o$79b2o$81bo13$2o$b2o96bo$o97b2o$98bobo!

It's an interesting approach. I wonder how lucky that p6 find was... Obviously a lot of objects will show up that will pretty nearly never appear in normal soup census results; I suppose a separate database will be needed to store symmetrical stuff in.

[knightlife]

dvgrn wrote:
I wonder how lucky that p6 find was...

I was using one of the features of Golly to "quickly" try the next pattern by using Shift-space to advance to a new initial pattern where the four outside gliders would be advanced one generation (by selecting all but the outside four gliders). When the result became uninteresting I would backtrack using "undo" and add four more gliders to the outside. If a computer search program could really tell what is interesting and what is not, it would get lucky pretty quickly! As we all know, it is pretty tough to program a computer to recognize what is interesting. I was monitoring intermediate results as well, to see unusual quad reactions near the center. I might not be able to quickly find another oscillator, but I will find unusual reactions that might lead to something else. A computer might miss an oscillator that is almost complete and just needs one more set of four gliders to clean it up.

Overall, I believe in this case the computer will get pretty lucky with a fairly simple algorithm.

dvgrn wrote:
suppose instead of starting with a random field...

I thought about this and would like to try very closely spaced still lifes only one cell apart but not touching each other. Some births would occur in the random jumble of still lifes creating unusual patterns initially. There might be something new there, just like the random quad salvos can create patterns you won't see (not true but extremely unlikely) in a uniform random soup. Adding symmetry to this should make it even more unusual relative to normal random soup. Perhaps letting the still lifes touch would make even more unusual results in short order (but less synthesizable).

As random soups of any kind evolve, their outcome in general starts to look very typical because the evolutions of sub-patterns that are different can merge and create the same result locally. Uniformly random soups have been tried for decades! They all tend to coagulate to the same stabilized patterns with only astronomically rare objects left to discover. It is time to try unusual "semi-random" soups.

[calcyman]

On the Soup Search homepage, why isn't the switch engine listed in the 'spaceships and puffers' section?

According to Achim's frequency tables, there should be several hundred switch engines in a search of this magnitude.

This is rather worrying. Perhaps the program can't recognise p288 puffers, or discounts infinite-growth patterns altogether?

[Nathaniel]

On the Soup Search homepage, why isn't the switch engine listed in the 'spaceships and puffers' section?

According to Achim's frequency tables, there should be several hundred switch engines in a search of this magnitude.

This is rather worrying. Perhaps the program can't recognise p288 puffers, or discounts infinite-growth patterns altogether?

Don't worry, they're being caught and there are indeed several hundred in the database; they're just not showing up publicly yet because of a bug I have yet to sort out that makes my "make the data nice" script not work on it. Not sure if it's because of the high period or what. I hope to have it fixed this week.

[knightlife]

knightlife wrote:
try very closely spaced still lifes only one cell apart but not touching each other

I tried and there is nothing special about it after a couple of generations. It is just another high density soup that dies quickly initially!

[Lewis]

I noticed quite a few of the patterns on the statistics page haven't got the names next to them or wiki links, even though the wiki pages exist.

And I might have already asked this but is there any way I or other forum members could help fix the data for objects not shown on the chart?