Page 12 of 16
Re: apgsearch v1.0
Posted: June 6th, 2015, 11:50 pm
by Rocknlol
Extrementhusiast wrote:Found a slight bug: when trying out a new rule for the first time, it briefly samples some soups from regular Life before switching to the new rule.
That's just apgsearch peer-reviewing b3s23/C1 hauls.
calcyman wrote:
New feature: Hauls uploaded to b3s23/C1 are now peer-reviewed by instances of apgsearch v1.1 (even those running different (rulestring, symmetry) combinations) to ensure the validity of the data.
Re: apgsearch v1.0
Posted: June 7th, 2015, 5:36 am
by apg
Rocknlol wrote:Extrementhusiast wrote:Found a slight bug: when trying out a new rule for the first time, it briefly samples some soups from regular Life before switching to the new rule.
That's just apgsearch peer-reviewing b3s23/C1 hauls.
Indeed:
http://tinyurl.com/ouvu2ub
Re: apgsearch v1.0
Posted: June 8th, 2015, 4:30 pm
by Extrementhusiast
It flashed by so quickly, I thought it was searching through entirely new soups, thus the confusion.
EDIT: Just for fun, I've calculated that there are exactly this many unique soups in C1 symmetry (counting rotations and reflections as the same soup):
Code: Select all
14 474 011 154 664 524 427 946 373 126 085 988 492 632 854 416 405 335 952 223 242 430 050 012 758 016
(that's 1.4474*10^76)
Re: apgsearch v1.0
Posted: June 10th, 2015, 1:22 pm
by gmc_nxtman
Why does apgsearch count these as a single still life?
Code: Select all
x = 32, y = 32, rule = B3/S0248
o3b2o4bo3bo2bo3bo4b2o3bo$2o2b2o4bob8obo4b2o2b2o$6b2o2bo2bo4bo2bo2b2o$
2bobo6bo2bo2bo2bo6bobo$4b2o7b2o2b2o7b2o$2b4obob4ob4ob4obob4o$2ob3ob2o
2b4o2b4o2b2ob3ob2o$8b4obo4bob4o$3bob6obob4obob6obo$3obob2o2bo3bo2bo3bo
2b2obob3o$o5bo6b6o6bo5bo$2o2b2obo3b2ob4ob2o3bob2o2b2o$obo2bobobo2bo6bo
2bobobo2bobo$2b6o2bobo2b2o2bobo2b6o$bo2bob2obo2b2ob2ob2o2bob2obo2bo$4o
2b3o5bo2bo5b3o2b4o$4o2b3o5bo2bo5b3o2b4o$bo2bob2obo2b2ob2ob2o2bob2obo2b
o$2b6o2bobo2b2o2bobo2b6o$obo2bobobo2bo6bo2bobobo2bobo$2o2b2obo3b2ob4ob
2o3bob2o2b2o$o5bo6b6o6bo5bo$3obob2o2bo3bo2bo3bo2b2obob3o$3bob6obob4obo
b6obo$8b4obo4bob4o$2ob3ob2o2b4o2b4o2b2ob3ob2o$2b4obob4ob4ob4obob4o$4b
2o7b2o2b2o7b2o$2bobo6bo2bo2bo2bo6bobo$6b2o2bo2bo4bo2bo2b2o$2o2b2o4bob
8obo4b2o2b2o$o3b2o4bo3bo2bo3bo4b2o3bo!
Re: apgsearch v1.0
Posted: June 10th, 2015, 8:01 pm
by gameoflifeboy
gmc_nxtman wrote:Why does apgsearch count these as a single still life?
Code: Select all
x = 32, y = 32, rule = B3/S0248
o3b2o4bo3bo2bo3bo4b2o3bo$2o2b2o4bob8obo4b2o2b2o$6b2o2bo2bo4bo2bo2b2o$
2bobo6bo2bo2bo2bo6bobo$4b2o7b2o2b2o7b2o$2b4obob4ob4ob4obob4o$2ob3ob2o
2b4o2b4o2b2ob3ob2o$8b4obo4bob4o$3bob6obob4obob6obo$3obob2o2bo3bo2bo3bo
2b2obob3o$o5bo6b6o6bo5bo$2o2b2obo3b2ob4ob2o3bob2o2b2o$obo2bobobo2bo6bo
2bobobo2bobo$2b6o2bobo2b2o2bobo2b6o$bo2bob2obo2b2ob2ob2o2bob2obo2bo$4o
2b3o5bo2bo5b3o2b4o$4o2b3o5bo2bo5b3o2b4o$bo2bob2obo2b2ob2ob2o2bob2obo2b
o$2b6o2bobo2b2o2bobo2b6o$obo2bobobo2bo6bo2bobobo2bobo$2o2b2obo3b2ob4ob
2o3bob2o2b2o$o5bo6b6o6bo5bo$3obob2o2bo3bo2bo3bo2b2obob3o$3bob6obob4obo
b6obo$8b4obo4bob4o$2ob3ob2o2b4o2b4o2b2ob3ob2o$2b4obob4ob4ob4obob4o$4b
2o7b2o2b2o7b2o$2bobo6bo2bo2bo2bo6bobo$6b2o2bo2bo4bo2bo2b2o$2o2b2o4bob
8obo4b2o2b2o$o3b2o4bo3bo2bo3bo4b2o3bo!
I'm not sure what you mean.
If you mean "why does apgsearch consider ob2obo$6o2$6o$ob2obo! a single still life":
Because one of them on its own is unstable:
Code: Select all
x = 6, y = 2, rule = B3/S0248
ob2obo$6o!
If you mean "why does apgsearch count both as a single still life":
It doesn't. The reason there are only three sample soups while the
object count is at six is because the object count gives the total number of occurrences for an object, not the number of soups in which it occured. In this case, since the still life occurs twice in each soup, the object count for xs20_rarrar is twice the number of soups that occur.
Re: apgsearch v1.0
Posted: June 10th, 2015, 8:05 pm
by gameoflifeboy
Speaking of pseudo still lives, did apgsearch ever learn to separate xs20_gbb8r2qkz11?
Re: apgsearch v1.0
Posted: June 30th, 2015, 12:24 pm
by gmc_nxtman
What does
this mean?
It says the haul was commited by calcyman, but "integrity-verified" by me.
Catagolue wrote:Haul
This haul was submitted by
apgoucher@gmx.com on 2015-06-30 at 12:39:05 UTC.
It contains 72826499 objects (285 distinct) generated from 3333333 soups.
The integrity of this haul was verified by gmc_nxtman.
Re: apgsearch v1.0
Posted: June 30th, 2015, 4:10 pm
by apg
It says the haul was commited by calcyman, but "integrity-verified" by me.
Your instance of apgsearch peer-reviewed one of my hauls.
Re: apgsearch v1.0
Posted: June 30th, 2015, 4:23 pm
by gmc_nxtman
Sorry if this sounds a bit silly, but.. why would it need to do that?
Also, how is it possible to just have massive amounts of hauls going at once? Why is biggiemac filling 80% of the pie chart? I have 6 instances of the search running 99% of the time...
Is anybody still running 8x32 searches? In my opinion, 8x32 is severely underrated....
Re: apgsearch v1.0
Posted: June 30th, 2015, 4:26 pm
by Kazyan
gmc_nxtman wrote:Sorry if this sounds a bit silly, but.. why would it need to do that?
It's possible to falsify data if that's not done.
Re: apgsearch v1.0
Posted: July 1st, 2015, 7:59 pm
by apg
Kazyan wrote:gmc_nxtman wrote:Sorry if this sounds a bit silly, but.. why would it need to do that?
It's possible to falsify data if that's not done.
Indeed. This is also useful for testing major changes to apgsearch to ensure that they produce accurate data.
For example, I'm writing a C++ version of apgsearch (b3s23/C1-only and highly optimised), which so far appears to be almost twice as fast as the Python version. It doesn't need a GUI installed on the machine, and can easily be executed from a bash script, so is ideal for machines designed for heavy computation. (In particular, Tom Rokicki has several such machines.)
gmc_nxtman wrote:Also, how is it possible to just have massive amounts of hauls going at once? Why is biggiemac filling 80% of the pie chart? I have 6 instances of the search running 99% of the time...
He had 25 quad-core computers, each core of which was running an instance of apgsearch, so 100 instances in total. These ran full-time for about 7 weeks, for a grand total of 120,000 CPU-hours of work.
Re: apgsearch v1.0
Posted: July 2nd, 2015, 10:17 am
by gmc_nxtman
For some reason
some of the hashsoups have "x = 32, y = 16" as oppose to "x = 32, y = 8". It's not a fatal or major bug but I just noticed it yesterday. I also think 4x64, 1x256, and 2x128 should be valid symmetry options.
Another thing that would be nice is giving soups with nothing at all in their census very high scores.
Re: apgsearch v1.0
Posted: July 4th, 2015, 7:42 pm
by gameoflifeboy
I hope there's a way for someone to delete comments on Catagolue, because many object pages have suddenly gotten a lot of spam, like Pentadecathlon, Pulsar, Blocker, Cis- and Trans-Queen Bee Shuttle, Fumarole, Boat-Tie, and so on.
It would probably be a good idea to disallow anonymous users from posting comments.
Re: apgsearch v1.0
Posted: July 5th, 2015, 5:41 am
by apg
I hope there's a way for someone to delete comments on Catagolue, because many object pages have suddenly gotten a lot of spam, like Pentadecathlon, Pulsar, Blocker, Cis- and Trans-Queen Bee Shuttle, Fumarole, Boat-Tie, and so on.
Thanks. I've manually deleted all the offending comments.
It now prevents any comments containing the substring 'href' (try it for yourself!), which seems to match 100% of the spam comments and 0% of the other comments.
Re: apgsearch v1.0
Posted: July 5th, 2015, 7:47 am
by Billabob
It now prevents any comments containing the substring 'href' (try it for yourself!), which seems to match 100% of the spam comments and 0% of the other comments.
I really regret trying now. That should teach spammers, though...!
Re: apgsearch v1.0
Posted: July 5th, 2015, 11:16 am
by gmc_nxtman
I think Noah's ark, and the other arks (I'm not sure how many, could somebody give me a list please?) should be named instead of the standard "yl1344", "yl12", etc. as "ark1", "ark2", ... "Noah's Ark" unless it is impossible to name all the arks due to their massive abundance as discovered recently.
Generally
more of the patterns should have names.
A few others:
Pulsar on pentadecathlon 1
Cuphook
Generally the cuphook variants
This variant of the stillater
Buckaroo with integral
Mold siamese snakes
Generally the mold variants
The trivial p6s
calycman wrote:(try it for yourself!)
Re: apgsearch v1.0
Posted: July 14th, 2015, 6:48 am
by Billabob
HTML works in comments? The
snake page crashes your browser after a few seconds.
Re: apgsearch v1.0
Posted: July 14th, 2015, 1:31 pm
by apg
Billabob wrote:HTML works in comments? The
snake page crashes your browser after a few seconds.
I'm disappointed; I was expecting a playable game of Snake. I've now deleted those comments.
Re: apgsearch v1.0
Posted: July 14th, 2015, 6:30 pm
by gameoflifeboy
What were the comments?
Re: apgsearch v1.0
Posted: July 15th, 2015, 4:02 am
by Billabob
calcyman wrote:I was expecting a playable game of Snake.
Check again.
gameoflifeboy wrote:What were the comments?
They crashed your browser.
Re: apgsearch v1.0
Posted: July 19th, 2015, 5:00 pm
by Alexey_Nigin
Re: apgsearch v1.0
Posted: July 19th, 2015, 8:45 pm
by gmc_nxtman
I have a rather incredible bug. Some of my soups in B368/S245/D8_1 are getting scores of over 24,720. What might be causing this? If this is fixed, I also think that there should be a "highest-scoring soups" section for every symmetry, so that I could give proof.
Re: apgsearch v1.0
Posted: July 20th, 2015, 12:28 am
by Extrementhusiast
And another cosmic ray event happened, this time with an LWSS that doesn't get analyzed even after a billion generations. Perhaps an in-script abort function could be added, so that I (or somebody else) don't lose another 800K soups?
Re: apgsearch v1.0
Posted: July 20th, 2015, 2:48 am
by apg
Regarding the bugs, there's a C++ version of apgsearch being tested at the moment. This was basically written from scratch, so shouldn't contain any of the original bugs present in the Python version. Also, it's 7 times faster.
I have a rather incredible bug. Some of my soups in B368/S245/D8_1 are getting scores of over 24,720.
That's understandable: scores are based on how frequent an object is in B3/S23, so a puffer creating alien objects (such as 'big blinkers') would cause huge scores.
Re: apgsearch v1.0
Posted: July 20th, 2015, 3:44 am
by simsim314
May I ask what iterator did you used? I would suggest to port the soup search to gpu as well, this should give another order of magnitude jump.