Solver recommends putting quality modules in crafting machines with fluid outputs

[sjdrodge]

An example way to reproduce is running the script with the following options: --input-items 'heavy-oil=0' 'scrap=0' --output-item quality-module-3 --module-cost 1 --productivity-research 'processing-unit=1.1' 'plastic-bar=0.9' 'scrap=1.1' --allow-byproducts --output-amount=0.016 --disallowed-crafting-machine biochamber

This will return a factory that recommends putting a quality module in e.g. Holmium Solution and Ice Melting Chemical plants.

[scottmsul]

Thanks for noting this. After looking into this I'm not convinced this is a bug and is actually the correct solution, though I'll admit I'm still not 100% sure. Basically I tried an output of holmium-plate instead which was still putting quality in holmium solution, then I checked again with a cost on scrap and then holmium solution was 3P as expected. I think allowing scrap to be free makes the solver not care if it wastes scrap since it would be the same cost either way. However when I went back to your example with a cost on scrap it was still putting quality in the holmium solution, though in the case of making an output item of quality modules there's a whole bunch of other items involved, so maybe it's possible that holmium wasn't the limiting factor and was therefore able to use a less efficient holmium production chain without sacrificing cost.

I'll note the way the solver works is it only checks combinations of prod/qual modules with every module slot filled, so it's not necessarily that the quality module "helps" but rather that having 3 prods "isn't necessary" and 2 prod is "good enough" to reach optimal, and the third slot just has a quality as a "filler".

[scottmsul]

Thinking about it even more, since productivity modules slow the machine down by 15% and quality modules only slow it down by 5%, if the extra productivity isn't needed relative to other ingredients then in some cases using prod modules instead of quality modules could make the cost go up. It would be better to leave an empty module slot but the solver isn't checking this case.

[sjdrodge]

Thanks for taking a look!

I haven't examined the code yet, so I apologize for my ignorance, but does this mean that the script can only recommend either unmoduled or fully moduled machines?

I also noticed that the module count seems to be scaled by building count, and neither number rounds up, which can result in some rather amusing solutions where several unnecessary recipes are used to reduce the load on buildings that already have a count between 0 and 1. In practice this has the opposite of the intended effect, and instead increases the amount of buildings/modules needed to actually build the factory.

There is one huge caveat to the above "complaint," which is that the game now allows us to set recipes on buildings using the circuit network, so 1 isn't actually the minimum building count anymore. However, the script is of course still unaware of the more subtle rounding properties implied by the ability to share buildings between recipes.

And I'm sorry, I've now surely strayed into feature request territory, but it would be incredible if the script could round building/module count up, and/or understand some machine sharing (especially recyclers, which are often trivial to share).

That said, I understand the main purpose of the script is to optimize resource consumption for a given output, and it does a great job of doing that! Thanks for your work!

[scottmsul]

I haven't examined the code yet, so I apologize for my ignorance, but does this mean that the script can only recommend either unmoduled or fully moduled machines?

I'm using google's ortools for the solver. I had assumed that solutions had to be real-valued but it looks like you can make "int vars", that would be an interesting setting. Working on some other stuff right now but I would like to play around with this idea at some point.

I also noticed that the module count seems to be scaled by building count, and neither number rounds up, which can result in some rather amusing solutions where several unnecessary recipes are used to reduce the load on buildings that already have a count between 0 and 1. In practice this has the opposite of the intended effect, and instead increases the amount of buildings/modules needed to actually build the factory.

The idea for the "module count" is to help convey how "expensive" a factory is to setup. The more I've thought about this the more I'm leaning towards "building count" being the superior cost variable, but it's somewhat subjective.

There is one huge caveat to the above "complaint," which is that the game now allows us to set recipes on buildings using the circuit network, so 1 isn't actually the minimum building count anymore. However, the script is of course still unaware of the more subtle rounding properties implied by the ability to share buildings between recipes.

The idea of recipe switching is a good point I hadn't considered. I think it probably wouldn't matter much in practice if you're trying to make a large amount of something, as in that case some of the steps would probably require many (>> 1) buildings to maintain throughput, and recipe switching would only save on the negligible number of buildings that only need less than one each.

And I'm sorry, I've now surely strayed into feature request territory, but it would be incredible if the script could round building/module count up, and/or understand some machine sharing (especially recyclers, which are often trivial to share).

Currently working on a better interface, stay tuned!

That said, I understand the main purpose of the script is to optimize resource consumption for a given output, and it does a great job of doing that! Thanks for your work!

Appreciate the kind thanks. Note that strictly speaking, optimizing for resource consumption is only one way it can be used (costs on inputs, no costs everywhere else). The more I've thought about this problem, the more I've come to realize that "outputs per input" is probably not the right metric to optimize, for instance by using speed beacons a factory can be made dramatically more compact at the expense of a little bit of wasted resources, which is trivial to setup with a few extra miners.

[sjdrodge]

Your last point would be an amazing feature for the optimizer to have! i.e. understanding that using prod modules instead of quality modules allows for speed beacon usage. I suppose it's even possible that using a limited amount of speed alongside quality could result in big building count savings.

[scottmsul]

Your last point would be an amazing feature for the optimizer to have!

Try running with --check-speed-modules 😃