UPS wars: smelting challenge (design competition)

[TBTerra2]

find the factorio exe, and open a cmd window in the folder

[quyxkh]

For my own information, so I can get a rough idea how my design stacks up without bugging you: where do you get the plates/hr figure from? Running the benchmark doesn't give it to me, do you check the production history manually or ? And as an idle-curiosity while-I've-got-you, where'd that 100/216 factor come from? Would you by chance be normalizing your best design to a score of 100?

[TBTerra2]

the 100/216 is because the simulation is for 100k ticks, but the production is for one hour (which is 216k ticks)

as for your score, its 27ish for the big one, normal values seem to be 35-40 with the best so far being 55. i suspect your striped down one will score higher due to not having a bloated giant logictics network

also if you have multiple different concepts, do send them in, although i can see why you would want to be able to refine a particular concept fully before sending it

for rough testing, take the number of plate per hour (from production stats) * 100/216 then divide by the time in ms that the simulation takes

[Optera]

Here's a small and a large version of the same design using direct insertion, bots and trains.
I'm really curious how well this design scales and competes against other designs. Though I'd never use a design wasting so much resource space in a normal map.

2017-08-14-08-09-26-1241785.jpg (919.42 KiB) Viewed 7324 times

[Xtrafresh]

Mine was pretty similar, except I rearranged the smelters and miners a bit for less wasted space:



I am getting a ~60% coverage from this setup, yours only mines 44% of the patch. The bright side though, is that you can pretty easily shift the whole design over to the right for a "second run" on the ore patch, getting an 83% coverage by the time that one depletes. Mine will do just over 90% after the second run, but it'll require a verticle shift, and thus re-laying beacons, roboports, rails, the works.

Still, pretty unsatisfying to leave a bit of ore laying around like that, haha.
My first submission used a simplified rail setup, I'm curious to see if that made a difference.

[Optera]

Yeah leaving just one tile untapped makes me crazy. I had instances where i mined a single tile with 4 drills with 3xt3 speed module just to clean that out.

Those little design changes should have no influence on UPS. Any noticeable difference will probably stem from different logistic network sizes or train system.

[Xtrafresh]

I've done that very same thing. I didn't have a good way to take the resulting or away, so it just went into a chest that I then shotgunned. But the ground is nice and smooth now

Hmm, my big one is a total disaster when it comes to the logistics network, I don't have much experience with bots (I usually build belt bases). My train system is crazy optimized though. three separated simple circles with two massive trains each. No pathing decisions, no crossings, hardly any signals. Tonight I'll have time to go address the problems.

[Optera]

My train system is crazy optimized though. three separated simple circles with two massive trains each. No pathing decisions, no crossings, hardly any signals. Tonight I'll have time to go address the problems.

Makes me wonder how much UPS train pathing actually takes.
It always seemed very minor to me considering my vast town style map runs smoother than a single megabase with similar production stats.

I'll change the rail layout of my submission and compare them outside of this challenge.

[Frightning]

What are the dimensions of the ore patch? (tilesxtiles) I need them for theoretical calculations so I can size my smelty and trains appropriately.

Addendum: A reasonably big throughput minimum would probably be wise, because otherwise smaller setups will almost assuredly win thanks to things with worse than linear big-O scaling. (e.g. bots are O(n^3) iirc)

[TBTerra2]

What are the dimensions of the ore patch? (tilesxtiles) I need them for theoretical calculations so I can size my smelty and trains appropriately.

Addendum: A reasonably big throughput minimum would probably be wise

the ore patch is 200x400 tiles

im probably going to impose a minimum of 50k/min and maximum of 500k/min. (the lower due to it simulating too fast to get a reliable result, the upper because it takes so long to test)

[Optera]

I hope you're going to post screenshots of all designs and their numbers once you're done.

[TBTerra2]

I hope you're going to post screenshots of all designs and their numbers once you're done.

that is the intention

the preliminary scores for the entrys i have tested are in the google doc, but im keeping builds under wraps to some degree to minimize idea theft

[Frightning]

I hope you're going to post screenshots of all designs and their numbers once you're done.

I would also like to propose making a scatter plot of the plates/min and UPS/plate numbers, because it would be interesting to see the overall shape of best scoring results.

[Xtrafresh]

Wow! So the bots are clearly murdering me. I know you said idea theft is frowned upon, but I no use trying to ride a dead horse. And looking at how I need to pretty much triple my score to have any success, I think we've proved that it is indeed a dead horse. I'll optimize a bit for academic purposes, but no bots seems the way to go. More files for the grinder tonight!

I hope you're going to post screenshots of all designs and their numbers once you're done.

I would also like to propose making a scatter plot of the plates/min and UPS/plate numbers, because it would be interesting to see the overall shape of best scoring results.

Should be easy enough to do yourself, once the final doc has been released?

[Frightning]

Wow! So the bots are clearly murdering me. I know you said idea theft is frowned upon, but I no use trying to ride a dead horse. And looking at how I need to pretty much triple my score to have any success, I think we've proved that it is indeed a dead horse. I'll optimize a bit for academic purposes, but no bots seems the way to go. More files for the grinder tonight!

I hope you're going to post screenshots of all designs and their numbers once you're done.

I would also like to propose making a scatter plot of the plates/min and UPS/plate numbers, because it would be interesting to see the overall shape of best scoring results.

Should be easy enough to do yourself, once the final doc has been released?

I suppose I could...also, regarding bot based designs...if I am correct in recalling the bots being O(n^3), then you may be able to make significant gains by modularizing your logistics network, the mathematical reasoning for this being advantageous can be seen as follows:
Suppose we have a need of a logistics network of size n, if we simply build a single network then it takes roughly n^3 amount of work, but if we instead break it into k smaller networks (ideally of roughly the same size as each other), then we instead have k networks each of sizes n_1, ... , n_k, with n_1+...+n_k=n. Now, the modular network requires about Sum((n_i)^3) for i running from 1 to k, but the single network is n^3, which equals (n_1+...+n_k)^3. If you know your algebra well enough, you will realize pretty quick that this latter quantity is generally far larger than the earlier sum because not only do you have each (n_i)^3, but you also have all the 'mixed' terms, of the forms (n_i)^2(n_j) and (n_i)(n_j)(n_k). So your performance savings for modularizing your logistics network is roughly the value of those mixed terms, which can get pretty big.

Edit: A quick numerical demonstration: suppose n=10,000, and all the n_i=1000, and k=10. Then we see that a single network requires 10,000^3=1,000,000,000,000 (1 trillion), but the modular network requires, sum(1000^3) running from 1 to 10, or 10*(1000^3)=10,000,000,000 (10 billion) or 1/100th the processing power!

[Xtrafresh]

I don't know my algebra all that well to be honest, haven't used it in too long
But what you are saying makes intuitive sense to me also, I got a bit lazy when getting to the final "bot" stage of the design to be honest. Let me dive in and get things right

[TBTerra2]

quick update on testing.

Traintown MK2: there is so much buffering that i cant get a production reading (literally 4 full chests of ore/plate per furnace). you will need to eather limit the number of items per chest or run it on your pc and get it to back up, becasue im not waiting for it to when i have other designs to test

Optera V2: im going to double this up because it has production too low to get reliable performance readings (it should still get a good score though)

Hypnobunny: you are using way over the 360 void chests given to you and you dont need them, so ive changed your unloading to one side only, this will probably do nothing to production but will increase your score

Xtrafresh MK2: you've said there will be a mk2 but i haven't got anything from you

[Xtrafresh]

Here's v1.1, the slightly cleaned up version of my first submission. It won't be a winner, but it'll be interesting to see the difference now that the bot networks are cleaner. using two massive 6-30-12-30-6 trains to dump 240K copper per run. Reduced their stops from 5 to 2. Reduced the number of setups from 3 to 1, for cleanliness. Each setup has two split bot networks, with a lot less roboports and only 600 bots each, and I even brought the max number of rows down from 4 to 3, for shorter bot travel. It should be doing 83K/m, just on the safe side of your 50K minimum. Can easily blueprint and triple the setup though if you want to increase the load.

Mind you, this one is purely academic, as with all these bots I still won't win, not even close. I'm coming in later with a proper V2 that should do better, and also be a bit more usable in the real world.

[Optera]

Optera V2: im going to double this up because it has production too low to get reliable performance readings (it should still get a good score though)

When you do please don't change the logistic network size or number of bots in a module.
v 3.2 is using the 5 elongated v2 modules with more bots. Should 4 v2 modules perform much differently it will show how much logistic network size impacts performance. v2 (~400/800 active bots) v3.2(~800/2k active bots).

Edit:
For reference here's a design I actually use on a lot of maps. I expect it to perform much worse than direct insertion though.

[quyxkh]

Well, let's see how this stacks up. I expect it'll be middle-of-the-pack, it feeds ore directly into smelters and loads the plates directly onto the train from there . . . hmmm. I think I know how to make it faster, but here's my frist cut at it:

quyxkh Smelting Eff 1 loop.zip

one loop unloading smelters direct to the trains

(2.31 MiB) Downloaded 141 times