[Discussion, not a guide] Best autoloader layouts for APS turrets

[boreal furnace]

Pretty ridiculous tbh

[hot nova]

worst comes to worst I can do a jank and literally shove the release version into it
C# lets you include resource files

some extra logic to switch which one depending on OS

[boreal furnace]

Best get to your implementation of the Traveling Salesman problem

[hot nova]

also I think I figured out how I might get some form of 5-clip
I could have the cooler cell type have no collision with itself (but still with normal pieces)
with that and a bit of extra to make grouping not instantly break it would probably work

[boreal furnace]

iirc it was mman.h that was the issue for me, might need to use this https://code.google.com/archive/p/mman-win32/

[stoic yarrow]

you should probably make the turret well area black and the surrounding white, cuz with it like that, it looks well... hard to actually see

[hot nova]

well it took me quite a while but I managed to get cell types and 5 clip working (at least I'm pretty sure)

[neon cairn]

This is not effective though

You want to have 2 autoloaders per connector

[hot nova]

coolers are set as self-intersecting so it results in the same thing

[neon cairn]

Yes but theres more gaps in it than what you want

^

you want something like this

[hot nova]

that's not optimal for 15x15
32 loaders is 2 loaders less than the optimal solution (34) that the solver finds

[neon cairn]

hmmm

interesting

[smoky fiber]

What even is 5-clip?

cuz it doesn't look like it has 5 clips?

[neon cairn]

Okay so

1 autoloader, 5 clips and 6 intakes

They could be rearranged to have the same shape as a 3clip 2m autoloader setup

[smoky fiber]

how tho, like is this a 1m auto specific thing?

[neon cairn]

Yes

[smoky fiber]

could you send a blueprint or screenshot in the chat?

[neon cairn]

Oh boy

Give me a sec then

@smoky fiber

This should work

[smoky fiber]

that's an... interesting loudout. Is it good?

[neon cairn]

Its the most efficient for 1m auto

If done right

[winter echo]

Brrt

[hot nova]

finally managed to fix symmetry for 5-clip
turns out it wasn't accounting for cell type when checking for symmetry groups so they were all self-conflicting

[hot nova]

I think I'm probably going to omit the shape editor for initial release and instead just get the export to prefab working

[boreal furnace]

Realistically, you've covered 99% of use cases

Shape editor is just nice for future-proofing or if people want to try other system Tetrises

[boreal furnace]

Export to blueprint is absolutely cracked if it works. Also would request different size and shape center holes in the area generator for turret-ception

[hot nova]

update: blueprints are a pain
specifically getting the right blockdata string is proving difficult
pretty sure I have everything else

BEHOLD THE FIRST GENERATED PREFAB (yes it's wrong I'm aware)

ITS WORKING

side note
for whatever reason the z? coordinate has 0 being the center instead of the edge
so to get it you have to do relZ - 0.5 * maxZ which is strange

[boreal furnace]

Awesome

[hot nova]

29x29 8m 3 clip export

5 clip is broken but I'll fix that tmr

[torn geyser]

amazing

[hot nova]

Fixes the last bugs for export I think, currently prepping the repo for release - might have it done today

[boreal furnace]

can export have an option to attach ejectors

[hot nova]

future version
currently I'm trying to figure out how to make the windows release build not commit sudoku

for some reason it hates existing

[hot nova]

yeah for some reason only the windows release build dies for certain UI updates
something weird about CoreMessagingXP dying
I'll see if I can figure it out tmr

[boreal furnace]

🇫

[hot nova]

well it still hates me but I did figure out how to get something working
at least I can release a non self-contained version

[hot nova]

https://github.com/trk20/APS-Optimizer

wasn't able to get a self-contained version working because windows really do be like that
still works though

LMK if you have issues running

[boreal furnace]

If you figure out how to accurately estimate how long it will take, please consider making an issue with the exact details or submit a pull request.

I think this might literally be a CS PhD project in and of itself lol

I'll see if I can get it running, probably tomorrow night

[cunning geyser]

grid width is 14 instead of 15 wide. also seems to be causing the thing ur using for the circle to spaz out (I think its just overflowing into the next row cuz its expecting it to be 15 wide not 15)

also its really small on 1440p (default 150% scaling). maybe add a scaler option cuz this wouldn't be usable on higher resolutions/smaller screens.

height is fine tho

i think i just entered 1416 (on accident) into the width and now its dying

yeah manual resizing and shaping doesnt fix it

(just using it not fullscreen now, why its sized different)

[hot nova]

Oops I guess I didn't check with different scaling and resolutions

[hot nova]

Yup that would definitely do it
I should probably check the entered dimensions before accepting

[hot nova]

https://github.com/trk20/APS-Optimizer/releases/tag/v1.0.1
Fixed the UI scaling for 150% and 175% plus added an option to increase the grid scaling

remember to get cryptominisat5.exe from their releases page and drop it into the same folder as mentioned in the installation instructions

[cunning geyser]

I never put the minisat file into the folder, so i did that now the solver outputs a result now. whoops!
The improper shaping bug is still there though (not that you said you fixed). now that I am getting an output it seems that this may be a display/input issue only, as the output is warped the same way (implying that it isn't warped under the hood). I havent gotten a chance to spawn the output ingame though, so heres what it output if you cant reproduce the bug.

[hot nova]

Just to check
windows desktop scale 150%, UI scale 2, 15x15 grid, and you're on the latest release

[cunning geyser]

Yes

Windows 10

[hot nova]

I'll check again, was pretty sure I fixed the UI scale but might need to add some extra tolerance

[cunning geyser]

I haven’t tested if it works at 100%

I think the issue is how it’s displaying the values, I’m guessing you are storing the grids as one long 1d array/list, or at least that’s how you are displaying them.

And it’s displaying each element in a line until it runs out of space, then looping to the next line

[smoky fiber]

I got updated to windows 11 by my dad, windows 10 superiority, never update, W11 is bad😭

All the UI for W11 is worse(IMO).

[cunning geyser]

Ye I don’t plan on going to win11

[boreal furnace]

I count 16x14 in the actual grid. Weird

Er.. 15x14

[cunning geyser]

thats been the issue

[hot nova]

figured out what the problem was

only happens if it's launched with 150% scale and not if it's launched at 125 then changed to 150 (which is how I was testing it)

[cunning geyser]

i just tested changing it to 100% (launched in 150%), and it went tonormal when switched to 100, and then messed up when switched back to 150%

wth

it fixes itself at 200%

[hot nova]

yeah

rounding differences

[cunning geyser]

rounding error lol

[boreal furnace]

mb, I've only been halfway paying attention, thought it was just indexing wrong 💀

[cunning geyser]

oh god changing the ui scaling also cooks it

kinda, i think a part of why this is happening is that the rowlength is being assigned based on the width of the viewport thing/width of boxes instead of the actual rowlength.

so throwing floats into the equation (via ui scaling to non-multiples of 100) borks it by making it slightly too small to fit the number of boxes it needs to.

thats the only wayh i can explain the rows overflowing into the next row

[hot nova]

https://github.com/trk20/APS-Optimizer/releases/tag/v1.0.2 hopefully it's fixed for good now

[cunning geyser]

seems to work

also was I right?

perfectly optimized 💯

[hot nova]

oh god mixed 3 and 5 clip why

[hot nova]

basically the UI framework doesn't have a proper resizable 2d grid so if it doesn't have enough size for a row it wraps around
I just had to add a few pixels of margin so that doesn't happen with the scale rounding errors

[cunning geyser]

why does 13x13 always work so well

even 5 clip tetris looks cool on 13x13

[hot nova]

It just does
Try the symmetry options, usually hard vertical symmetry works best for 5 clip

[cunning geyser]

woah its even better on 15x15

damn it can find really high fill % on 5 clip compared to 4 clip

what da hell turning on symmetry gave more clusters

why did i just try to do rotational syymmetry with 5 clip

damn it actually worked really well

it would be useful to have a counter for how many single cooler lines there are (ones with only 1 cluster instead of 2 attatched)

so it would be easier to compare different tetrises

[hot nova]

there was a number of hours spent trying to get 5-clip to not break all the other logic
glad someone is actually using it lol

[cunning geyser]

heh, yeah it looks complicated

funny how on 15x15 it works best on rotational 180 hard symmetry

8 unpaired clusters

[hot nova]

the way it actually works is that it allows coolers to self-intersect, which means for volume filling it will pretty much always go for the maximum amount of paired clusters

also why it takes longer to solve compared to 3/4 clip usually
it can't assume the number of cells filled is a multiple of the LCD of enabled shapes

something like that

3 clip will pretty much always fill the space best

[cunning geyser]

couple suggestions, then i go back to studying,1. unpaired cluster count, 2. when its done, have it rotate the unpaired clusters so the cooler is in contact with an empty block (so components can be attatched). i'm not sure how hard that would be but it would be nice

[hot nova]

The answer for 2 is hard

It would have to do a whole other solver loop with a different problem deconstruction

[cunning geyser]

oof

[hot nova]

One plus side to 5 clip is that you can stack it (upwards) and export supports that

[cunning geyser]

another 180 sym w

180 deg symmetry might be the best one

[hot nova]

Yeah 4 clip generally does best with rotational symmetry

3 clip is a toss up

[cunning geyser]

ngl 5 clip is defo faster than 4 clip

2.5 mins for 23x23 5 clip but im on 8 mins for 4 clip and ist still not done

didn't do rotational symmetry though that might've helped

930 and only attempt 12 this is gonna take too long

[hot nova]

Hard symmetry makes the solution space much smaller

[cunning geyser]

oh

it finished before i could find the exe in taskmanager

[hot nova]

Especially for more sparse layouts like 4 clip

[hot nova]

That looks pretty nice

[cunning geyser]

doing 4 clip rotational symetry now

looks optimized to me

[hot nova]

yeah it leaves 1 cpu core so the UI doesn't have a stroke

[cunning geyser]

You run the app file

It should be called optimizer v3

hmm, seems like a weird name considering it does

Should probably be called aps generator or something

[cunning geyser]

wth are you doing with extentions

double click that

its an application, It sounds like your trying to run it like a peice of code or something

source code?

[cunning geyser]

15x15 armored

[hot nova]

Sorta leftover from the initial name I gave it lol

This is the third time I try to tackle the same problem, and once I got the actual core logic working I made a UI for it, never bothered changing the name to something that made more sense

[boreal furnace]

Finally got to test it, worked flawlessly for me

5-clip is very neat

[hot nova]

Glad to hear it
Was an absolute pain to implement lmao

[boreal furnace]

export to blueprint was almost painless, only feedback would be letting you set a default path / have it remember the last path so it is literally one or two clicks instead of 4 or 5

but tbh that's a crazy feature already

looking forward to when it also does ammo inputs and ejectors :^)

[cunning geyser]

god damn this thing is fast, solved a 25x25 in less than 2 minutes

[hot nova]

What on earth were you doing to make 25x25 take more than 1?

5 clip soft 4 way symmetry maybe?

could just be differences across cpus actually but still wouldn't expect close to 2mins for 25^2

[cunning geyser]

laptop

it was like 90 seconds

R9 5900H

i think it was 5 clip rotational 180 or vertical. also im doing 25x24 to get the symmetries to behave naturally for a front armored turret

took 1:50 for 4 clip rotati9onal 180 hard

[cunning geyser]

ngl ive found that 4 clip tends to take the longest

heh, its different every time you do it wwith these settings

this one gernarates super rarely with soft settings. i think i got it with hard too at one point but im not sure

[hot nova]

huh must be CPU - I'm getting ~40s for 4-clip 25x24 180 degree hard symmetry
7800x3d go brrr I guess

[cunning geyser]

bro, why would you not expect that to do anything😂
of course it's gonna be faster on top of the line hardware

you have 16 cores and +40% faster cores

[hot nova]

*8 cores but yeah I get your point
crypyominisat probably favors the large cache size as well

[boreal furnace]

@hot nova something I thought of recently - have you considered beltfed tetris yet?

Would be something like choosing 2-6 ammo intakes, but for those less than 6 you would have permutations of intake placement

[hot nova]

I've never really seen beltfed tetris so idk how that it's supposed to look

[boreal furnace]

It's one autoloader + one clip, then fitting as many inputs on that as you can. Most people just do like a 5x5 windmill thing with it

[dry sleet]

sorry, I have stopped playing as much FtD as I used to, and haven´t really been keeping up with the chat

is it doing well?

[boreal furnace]

Trk finished his solver, posted over in the other thread, handles basically 99% of use cases, has neat features like export to prefab, and the biggest pro - it isn't in MATLAB

[dry sleet]

so it´s a complete ship of Theseus

no parts of my original code remain xd, prob for the better given my level of optimization xd

what size have you managed?

[boreal furnace]

I've not really pushed it, but I know the SAT solvers can get into the 30s for single solutions before they start to chug a bit

What is funniest to me about this project is that even before the solvers started coming out, basically all the practical sizes were found

That being said, I think the 5-clip tetris in the new solver is really neat and actually better than what people came up with prior

[boreal furnace]

And also being able to specify symmetry is nice too

[hot nova]

33x33 180 rotational symmetry

[dry sleet]

13x13 and 15x15 are practical enough I think xd

[hot nova]

recently made a 21x21 with it

probably going to end up on multiple future designs of mine because I like large guns

[hot nova]

probably going to look at implementing ejector/intake placement at least for the bottom sometime this week

[marble rivet]

no, 2 input for max density and 4 input for max cost efficiency

[blazing cloak]

Sorry if this is off topic but Is there a reason horizontal layouts arent discussed more often? This design fits 4 meter clips into a 13x12 foot print that is only 3 meters tall. Theoretically this layout is both more efficient for space and material because the opposed autoloaders essentially halve the length of the guage snake.

[torn geyser]

Probably because vertical layouts are a lot more modular/easy and fast to design the exact gun you want whereas horizontal requires unique tetris for most combinations

[midnight egret]

You’re never going to have a gun with such low cooling / recoil / rail requirements that that could amount to any savings

They’re not common because it’s not very convenient. The shape’s more complex and the space more limited.

due to the low amount of granularity you may also end up with a decent amount of unused space which people might not like

[fair forum]

one possible advantage is that a piercing shot could take out only a single loader cluster instead of a whole line, but they're much less convienient to fit and don't really have reusable layouts

[stoic yarrow]

also, (from what i know) its rare for you to have a space that is that wide but shallow like that. so you'd have to stack this slab ontop of itself, wich would be less space efficient due to the ammount of extra coolers required for connecting when building in smaller spaces, wich would be less when compared to regular tetris. the holes that are gonna form are also more difficult to fill with stuff, and would be more difficult to make, with not many benefits.

[coarse rampart]

Just realised the 3 clip solver could be used for single-input laser tetris as well

[neon cairn]

Kinda yeah

[boreal furnace]

At some point trk will code user-definable pieces and then you can use it for pretty much any tetris that is simplifiable to 2D

Ofc that is very limited still

[hot nova]

the actual work for the solver allowing user-definable shapes is already mostly done - it's just the export mappings and UI that would be tricky

cell types themselves are also very generic

I think I saw something about someone exporting all the block types - if I figured out how rotation was encoded instead of just copying the value blindly I could theoretically end up supporting anything, although unsure if anyone would use that

[snow basin]

I am trying to use the APS optimizer and I am getting this error. What have I done wrong?

[coarse rampart]

Read the README.md file on github.
This will tell you that that part must be downloaded separately, and where to download it from.

@snow basin

[snow basin]

ok thanks

[hot nova]

Good news: I'm finishing up the implementation of allowing a generated bottom layer of ejectors and ammo intakes for the 3 and 4 clip exports, will probably make a prerelease later today

[hot nova]

https://github.com/trk20/APS-Optimizer/releases/tag/v1.0.3 pretty sure it works, LMK if there's any issues with the new export option

[hot nova]

holy hell I may have finally figured out how to have it not be annoying with .net runtime

[hot nova]

https://github.com/trk20/APS-Optimizer/releases/tag/v1.0.4
ok, now bundled correctly (I think) - also now auto-downloads cryptominisat if it's not there already

[fair forum]

cryptominisat sounds so much like a crypto mining library lol

next goal: optimize the code so much you finish almost instantly and mine cryptocurrency for a few minutes before returning

[boreal furnace]

You get APS Tetris, @hot nova gets to use your GPU for sweet sweet Bitcoin, check the EULA

[coarse rampart]

still a pretty sweet deal

[hot nova]

https://tenor.com/view/todd-howard-it-just-works-bethesda-this-all-just-works-gif-20598651

[hot nova]

Just had a thought: the code is already in C# - maybe I could eventually make this into an actual FTD mod
Although I haven't made a mod yet so not sure how hard that'd be

[hot nova]

anybody running linux that can help me test a linux compatible version of the solver?
VirtualBox (which I usually use for testing on linux) doesn't work because it doesn't expose the right CPU instruction sets for cryptominisat

[torn geyser]

sure

Looks good

[hot nova]

Yay!

[hot nova]

Thanks!

[worthy badge]

note

onedrive ALSO fucks with this

[hot nova]

onedrive fucks with everything I'm not surprised

[worthy badge]

yeah just an FYI to make sure when you download it you KEEP IT ON YOUR PC

[abstract lily]

lowk taking notes

im assuming this is for adding as much power for aps

[abstract lily]

how would someone launch this on linux?

[torn geyser]

it should just work

[abstract lily]

what app did you use for it?

im still somewhat new to linux

[torn geyser]

Wdym what app? This discussion is about trk20's aps optimiser program, the latest version is available here https://github.com/trk20/APS-Optimizer/releases/tag/1.0.7-pre

[abstract lily]

just saw the linux installation line

mb prob got an outdated version

[abstract lily]

ok got it to work

[abstract lily]

i belive in you

[abstract lily]

why did i make my turret 31x31 man its taking like 2 years

[fair forum]

is it still running? I don't think it's going to finish

[abstract lily]

i ended up putting symettry to quadrant for it to finish faster

[hot nova]

In almost all cases symmetry will make it faster to solve and be just as optimal, I personally don't use it without - usually hard vertical symmetry

As a sidenote: I have absolutely no clue how to estimate how long it takes, I've tried

[abstract lily]

now that i think about it

wouldnt it be possible to do something like this (or atleast a simplified version of this) for cram too?

it wouldnt be the most optimal but its good enough for something to start on

[dense comet]

It's called basic 3d cram.