"Hashwar": A Battle Royale for individual GOL configurations

[hash0]

I am opening the Alpha version of the new engine to the public:
https://lifehashes.net/CCC/Y206/

Upon loading the website, you have to select two Glyphs to compete by clicking the two "[LOAD]" buttons.

I find the "256 Rounds" mode (which is sped up) especially useful to gauge statistical properties of these Glyphs.

Next feature to implement will be an option to hash/mine and submit your own Glyphs (including RLE support).

[bobinou]

Quick update on the best runs found so far:
After 150B runs, the best non-repeating is at 2154.

Code: Select all

x = 16, y = 16, rule = B3/S23:T16,16
obobobobo3b2o$o2b5o2b6o$2o2b4o4b2o$o3b2o2bo3b3o$4ob2o2b2o4bo$3bob4o2bo
3bo$2o2b2ob4ob4o$bob2o2b2o$3bo4b2ob3o$ob2o2bo2bobob2o$bo2b2ob3ob2ob2o$
obob2ob2o2b2obo$5o3bo3bob2o$2o3bobo2b5o$3b4o2bob2obo$5b3obo2b3o!

and the best agar pattern so far is 5456

Code: Select all

x = 16, y = 16, rule = B3/S23:T16,16
b2obo3b4obo$4bo2bo3b2obo$2obobo2b2o2b2obo$b3ob5o4bo$bobob2ob3o4bo$3ob2
ob4o3b2o$2o2bo2b5obobo$2b5ob2o2b3o$obob2ob2o3bobo$3ob2obo3b3o$obo3b2o3
bo2bo$o3b2o2b2obo2bo$b4obo3bob3o$3bob2o2bobo$3o3b3ob4o$2ob2obo2bo4bo!

It is building up a neat database over time.

(https://imgshare.cc/rm85dpub)
A overview of the data:
https://hallucinat.in/details/overview
Also, all data is freely downloadable through a simple API:
https://hallucinat.in/api-docs

Edit:
Just got a even better one 2232

Code: Select all

x = 16, y = 16, rule = B3/S23:T16,16
2b2ob6ob3o$2ob2obobob2o3bo$b3obobo2b3obo$4o3b4o2b2o$2b2obobo3b2obo$o3b
3obo5bo$bob5o3bob3o$3o2b2o2b3ob3o$bo2bo5b2ob3o$b3ob2obob2ob3o$ob6ob6o$
b2o2b2o2bo5bo$2b3o4b2obob2o$obo2b4obo2b3o$obob3o2b2o2b3o$7ob4o!

[bobinou]

I am opening the Alpha version of the new engine to the public:
https://lifehashes.net/CCC/Y206/

Upon loading the website, you have to select two Glyphs to compete by clicking the two "[LOAD]" buttons.

I find the "256 Rounds" mode (which is sped up) especially useful to gauge statistical properties of these Glyphs.

Next feature to implement will be an option to hash/mine and submit your own Glyphs (including RLE support).

the visuals are amazing and its easy to use. love it!

[Vort]

After 150B runs, the best non-repeating is at 2154.

Nice result!
I wonder, however, why this pattern is so dense.
Did you tried to make searches with other, lower, densities?
If I remember correctly, 35% gives the best results.

[bobinou]

Nice result!
I wonder, however, why this pattern is so dense.
Did you tried to make searches with other, lower, densities?
If I remember correctly, 35% gives the best results.

Thank you, simply because I naively gave each field a 50% chance of being populated. But good point! I will give it a shot with 35%

[Vort]

Just got a even better one 2232

This result gives one more data point confirming my guess made several years ago:
With lifespan (excluding oscillation duration) going higher and higher, final state contains less and less "content" (oscillators and still lifes); top results produce just void.

[hash0]

It also confirms now that Nathan's original hypothesis from a while back ("lifespan on the torus is capped at roughly five times the torus size") has been thoroughly debunked.

[Vort]

It also confirms now that Nathan's original hypothesis from a while back ("lifespan on the torus is capped at roughly five times the torus size") has been thoroughly debunked.

80x80 result with lifespan of 171836799 by vilc debunks it even better.

[Vort]

Here are some ideas regarding torus size.
With small sizes (8x8, 16x16), random searches produce the best results.
With large sizes (64x64, 128x128), engineered solutions clearly dominate.
Also there is a range, around 32x32, where both solution types produce comparable results.
It may be fun to explore 32x32 size using different approaches (including mixed ones, where engineered solutions are augmented with randomness).
Such size allows to have single gun + several targets for it + random still lifes, but who knows what else can be invented. Maybe best random results can give hints about possible improvements for engineered versions.
Regrading methods of lifespan measurement: it is possible to have separate competition category for diehards (patterns with empty final state). Such results are easy to check: if LifeViewer says "Life ended at generation 4826", it means diehard pattern has lifespan of 4826.

[hash0]

That is a phantastic idea!

[bobinou]

The 10^12 visited hashes have been surpassed

[hash0]

Go, Bobinou! Shine on, you crazy diamond!

[hash0]

I'm hereby opening up the pre-release version to the esteemed public on this forum:
https://lifehashes.net/CCC/Y213/

All the currently registered Glyphs are there to select from for test matches (you have to select Glyphs by pressing the LOAD buttons on both sides - a menu will then open up to allow you to select the number of rounds to be played in the match).

All currently registered Glyphs can be seen here:
https://lifehashes.net/CCC/Y176/catalog ... C&limit=24

If you would like to register your Glyph discoveries, you can do so here (note that the input mask accepts both RLE and binary strings - AND automatically translates between the two):
https://lifehashes.net/GlyphRegistry/102/

(Sorry that all of these links are disparate - I am working on unifying all these different functions into a coherent whole that can be easily accessed on lifehashes.net).

[hash0]

I now have a dedicated twitch stream that shows the Hashwar engine in full tournament auto-battler mode:
https://www.twitch.tv/lifehashes