Page 1 of 1

Period recognition of linear growth in apgv5

Posted: July 20th, 2019, 2:28 am
by LaundryPizza03
I had previously noticed that the p42 guns for a glide-symmetric p8 ship in B2c3ai4rw5a/S23-a4ir was wrongly identified as yl84. One such gun is shown below:

Code: Select all

x = 8, y = 25, rule = B2c3ai4rw5a/S23-a4ir
4b3o4$4bo$4b2o$5b2o$4b2o10$4bo2bo4$o$o5bo$o4b3o$5bo!
Even more troublesome is B2ce3-ci5y/S2-c3-a, which has a p190, mod95 gun that fires p48 ships. This (quite common) gun registers as yl4560 and makes apgsearch very difficult, if at all possible:

Code: Select all

x = 7, y = 3, rule = B2ce3-ci5y/S2-c3-a
b2ob2o$o5bo$b2ob2o!
This seems to happen only with guns for higher-period ships. The following gun for a p2 ship correctly registers as yl17 in the rule's census:

Code: Select all

x = 4, y = 3, rule = B2c3aeny4a5i6i7e/S12a3ejry4etz5y6c8
o2bo$b2o$o2bo!
I have not encountered such difficulties with puffers.

Re: Period recognition of linear growth in apgv5

Posted: July 20th, 2019, 7:45 am
by Moosey
LaundryPizza03 wrote: I have not encountered such difficulties with puffers.
There's a puffer in highlife that produces the p10 so even though it has a fairly low period its apgcode is wrongly suggesting it has a period of several thousand gens.

Re: Period recognition of linear growth in apgv5

Posted: July 20th, 2019, 9:12 am
by dvgrn
LaundryPizza03 wrote:Even more troublesome is B2ce3-ci5y/S2-c3-a, which has a p190, mod95 gun that fires p48 ships. This (quite common) gun registers as yl4560 and makes apgsearch very difficult, if at all possible...
Run pop-plot.lua on 10,000 ticks of the yl4560 gun and I think you'll see why it gets identified the way it does:
yl4560.png
yl4560.png (43.86 KiB) Viewed 1862 times
95 * 48 = 4560. There's no way apgsearch can find population periodicity at lower periods if it doesn't exist. To figure out that this gun is really p190, apgsearch would have to do some very detailed and computationally expensive analysis of the actual pattern, instead of just keeping an eye on the rate of population change.

The opposite problem does happen occasionally, where the population changes are more regular than the pattern itself. Basically there are no promises that the "yl" number will match the period, or the mod -- they're just a reasonably efficient way of automatically generating a reliable identifier for the pattern.