Factorio is two games: pre and post robots.

[blazespinnaker]

My strategy:
It's important to build compact and fast. Electric miners asap before science. Collect all the stone for furnaces via Rocks. Walls are expensive so pipes 2-3 layers are recommended around 4-6 turret camps. Rush grenades for fast clearing of big attacks. Heavy armor + grenades + shotgun + fish for fast and unexpensive base clearing, if base is to near. Physical damage 3 and stronger explosives 1 are helpful. Physical damage 2 has no advantages against Small biters, though. Stay on yellow ammo. Use assembler2 and steel furnace to reduce pollution. Rush flamers with walls before spitters show up, bc Spitters are quite deadly for gun turrets.

Turrets spew iron like insane. It's very very crazy. That iron can be used to research oil/landmines/bots.

Try this map out, I can get to land mines in about an hour and a half, and likely much shorter if I optimize. It has oil nearby but the biter bases are very very close at start. There are some patches further out, but of course they have larger biter bases nearby.

>>>eNpjZGBkYGDQAGKGBnsQ5mBJzk/MgfFAmCs5v6AgtUg3vygVWZgz
uag0JVU3PxNVcWpeam6lblJiMVAxgwMEA03NLMrPQzeBtbgkPw9VpKQ
oNbUYWYS7tCgxL7M0F10vA+Nd2YNLG1rkGED4fz2Dwv//IAxkPQBaCs
IMjA1A24EqgWIwwJqck5mWxsCg4AjETiCDGBkZq0XWuT+smmLPCFGj5
wBlPICKHEiCiXjCGH4OOKVUYAwTJHOMweAzEgNiaQnQCqgqDgcEAyLZ
ApJkZPyY8bhj648vdox/Vn685JuUYM9Y1sz75aviFjugJDvIn0xwYtZ
MENgJ8woDkhMgUjftGc+eAYE39oysIB0iIMLBAkgc8GZmYBTgA7IW9A
AJBRkGmNPsYMaIODCmgcE3mE8ewxiX7dH9AQwIG5DhciDiBIgAWwh3G
SOE6dDvwOggD5OVRCgB6jdiQHZDCsKHJ2HWHkayH80hmBGB7A80ERUH
LNHABbIwBU68YIa7BhieF9hhPIf5DozMIAZI1RegGIQHTf9goyC0gAM
DMwMCfABGwd8btQBRL6Es<<<

IMHO it's a very nice classic deathworld map.

[blazespinnaker]

Maybe you try mods with tougher biters or declare self-limitations to dont use mines, flamers, efficiency modules, etc.

Yeah, can always create arbitrary hurdles of course. But I think it's pretty interesting to have a canonical rule set and compete on that. It'd be fun if folks were more collaborative (or co-epitive as they say), but that's a personal thing I guess.

[blazespinnaker]

I tried and I’ve found I am not a huge fan of deathworld marathon. It’s not really harder, but rather with expensive resource costs it drags everything out and slows the pace of the game. In some ways it dials down the biter difficulty as well. I got to mines on DW marathon and called it quits as I was getting bored. Once I am at mines the biters become an afterthought.

Personally, i see DW marathon as a slow Motion race against relativly even faster evolving biters. You are not so limited by placing speed and can react a bit faster. Also, you have to concentrate on really important researches. F.e. Heavy armor, fast inserter, etc. might not be that important.
I think it could be interesting to take the same biter evo speed to Progress speed relation as DW marathon and make it a bit faster. F.e. Only Double research and non expensive recipes, but more time and pollution evo and less attack consumption rate.

Ok, last post for now I found once you have land mines (and bots and efficiency 1 modules very soon after), any struggle is over, and it just becomes a marathon game, which I find is too slow paced. I'd rather be playing around with my megabase map.

[jodokus31]

Ok, last post for now I found once you have land mines (and bots and efficiency 1 modules very soon after), any struggle is over, and it just becomes a marathon game, which I find is too slow paced. I'd rather be playing around with my megabase map.

Yes agree. And i didnt even use mines last time. Flamers are insanly strong. And efficiency modules 1 are too good as first tier.

Try this map out, I can get to land mines in about an hour and a half, and likely much shorter if I optimize. It has oil nearby but the biter bases are very very close at start. There are some patches further out, but of course they have larger biter bases nearby.

IMHO it's a very nice classic deathworld map.

I love to maintain an automated base, which needs to be protected. Gun turrets is the only early option, i can think of. And i love railworld with DW biters and very far oil, which I have to search with car and get back via train.
I will have a look, when i can play again.

[jodokus31]

I'll prove this all out eventually, I guess. Getting world records isn't my thing, but the math backs it up. The speedrunners are too focused on keyboarding and memorizing sequences to do the math unfortunately.

I know, the speedrunners do the math pretty accurate.
The practice would show, how your strat will perform. You can slow down the game and pause, if you want to check theoretical gameplay as you dont need to hurry that much.

[blazespinnaker]

Hey that's a brilliant idea though rather obvious after the fact (as most brilliant ideas are). I will do that. Thanks for the suggestion!

However, I have a simple example of how speedrunners are making obvious mistakes. Look how they over producer miners before going to assemblers. It doesn't make sense to make more miners than what you can hand craft. Such over production is the antithesis of Just in time manufacturing.

I mean, look at this -

Screenshot 2020-12-07 194233.png (748.56 KiB) Viewed 3834 times

Those furnaces placed don't get miners for 3 minutes in this run. Not a huge impact, but an obvious mistake

And then go onto watch how many more iron/copper miners he creates even though he can't hand craft what they produce. Other runners do the same thing.

You should only make about ~12 miner/furnace pairs before going for assemblers. 12 furnaces produce 12 resources every 3 seconds, or 2 per 1/2 second which is the max you can hand craft. That's assuming you're crafting just gears. If you add copper to that you should produce even less miners. If you're crafting furnaces, you need a few more.

What speedrunners are good at is keyboarding and often you tend to overly weight at what you're good at which is why they are ignoring the math.


I will say Gotlap runners at least are prioritizing tech. The any% speedrunners are completely off base. Robots explode the search space for optimization and should be priority #1. Once you get to red engines/frames, you start building those in prep for construction bots.

One thing to keep in mind is that you need to keep all placed assemblers at 100% working productivity. As soon as you hit that - more miners.

Belts/inserters are overhead and should only be placed if you are doing non productive things not worth their ROI and they can alleviate that.

What are non productive activities? Walking around and hand feeding. Just walking around is terrible, hand feeding has some payoff.

Walking and hand feeding is the same thing as hand feeding and if you can optimize your map so that you're doing both, that's pretty good. Good speedrunners will do this. Better even is walking and placing.

Hand pulling is relatively productive early on as it keeps your hand craft queue busy. Having an inventory with a lot of items in it is an example of unproductive activity.

The most productive activity you can do is placing miners/assemblers/furnaces that will be used efficiently. Placing them when they will not be used efficiently is anti-productive.

[jodokus31]

Hey that's a brilliant idea though rather obvious after the fact (as most brilliant ideas are). I will do that. Thanks for the suggestion!

However, I have a simple example of how speedrunners are making obvious mistakes. Look how they over producer miners before going to assemblers. It doesn't make sense to make more miners than what you can hand craft. Such over production is the antithesis of Just in time manufacturing.

Overproduction is buffering, which is not very productive, but you need some leeway. F.e if you later place a copperwire assembler it's good to have a bunch of copper to not have to feed it constantly just in time.

I mean, look at this -

Screenshot 2020-12-07 194233.png

Those furnaces placed don't get miners for 3 minutes in this run. Not a huge impact, but an obvious mistake

And then go onto watch how many more iron/copper miners he creates even though he can't hand craft what they produce. Other runners do the same thing.

You should only make about ~12 miner/furnace pairs before going for assemblers. 12 furnaces produce 12 resources every 3 seconds, or 2 per 1/2 second which is the max you can hand craft. That's assuming you're crafting just gears. If you add copper to that you should produce even less miners. If you're crafting furnaces, you need a few more.

What speedrunners are good at is keyboarding and often you tend to overly weight at what you're good at which is why they are ignoring the math.

You always have to optimize, what is the limiting factor. I think, it's quite early the placing and interaction speed and not resources. Later handcrafting is not limiting anymore.
I dont want to say, that it's not important what you observed, but it's quite tough to make a plan and do it perfectly as human over long time.
If you rush bots asap, you should ensure you can use them properly. do you need a "proper" base and mall? Do you need electric miners before, or can you stay with burners till bots? If you have no blueprints, what will you build. Creating a blueprint from scratch is almost so much work like building it. Bots are quite chaotic with building. You need to have strategy that works reliable.
I mean the speedrunners are experts in what they are doing and it's working. Unfortunately, it doesn't help to claim that it could be performed better without proof. And it needs to be possible for a human, except you create a TAS (beware, the work is insane to do it)

[blazespinnaker]

I noticed another thing that the GOTLAP runners (and any% runners too, but they are generally misguided are missing, which is actually rather massive.

GOTLAP should be producing way more stone than they are in the beginning. The new guy at the top is sort of figuring it out, but he should be producing much more stone. When automation comes, he should drop down about 6 assemblers, instantly turn all the stone to furnaces via assemblers, and then repurpose the assemblers as gears assemblers, create a lot of gears, and then repurpose 4 or 5 of them as burner assemblers. Do this all via handfeeding. You need to build juggle the assemblers a bit more back and forth as you bring in more stone.

With this technique in about 120 seconds you can literally spam the entire patch with burners and furnaces, creating rather incredible amounts of ore production. Importantly, you also want over produce coal as you want to avoid having to have to go back and forth and refill the furnaces/miners, since you have so many of them now.

I did a quick prototype and I was pretty amazed at how much production I could generate so quickly. I had 1k/m iron plates per minute at 10 minute mark. For GOTLAP you need 3102 iron plates total for research. (Not including assemblers/labs). Finding a good map via preview that has coal/stone/iron/copper all meeting together to minimize walking distance is important here.

I imagine I could generate more by dropping some assemblers to create coil/gc so that I could create more assemblers. Burner assembly is a bottleneck and I imagine you want about 10 assemblers working on making these. The beauty of these assemblers is that you can repurpose them for anything once you're done with the burners.

The reason the runners miss this technique is they are overly focused on building cool looking factories with dazzling keyboard dexterity. The GOTLAP folks are starting to clue in with the hand feeding, but there is a lot more to learn.

[Koub]

Why don't you show all these people how your strategy is the best by doing your own speedruns ?
To be honest, right now, I have the feeling of listening someone slouched on the sofa watching pro football players and criticizing their poor skills.
I'm not saying you're one, but I'm sure you get the picture.

We have a saying in French : "La critique est aisée mais l'art est difficile", which roughly translates into "criticism is easy but art is difficult". Judging is way easier than doing.

As one great philosopher said

[blazespinnaker]

Nah, I'm in it for the math. Plus, all due credit, these folks have talent I do not. All I'd end up doing is creating demo videos for them to copy and fine tune and do with far more dexterity than I have.

I tell you what though, rather than resorting to a cheap ad hominem, why not address the facts of my argument with real facts? Or better yet, try it yourself. It's pretty easy. You'll quickly see that I'm right.

[Koub]

It's sad tbh. I like the kind of revolutions like what Fosbury did with his weird jumping style back in the time.

[blazespinnaker]

Or even better, why not contribute some interesting ideas to this thread rather than hijack it? How and why would you speedrun? collaboration is a fun thing to do, trust me.

I'm definitely opened to be told I'm wrong, if the argument is sound. That slowing down the game idea above was brilliant.


Also, just curious, when you replied here - did you bother reading my OP? Or did you just want to drop in and ad your two cents without reading it? There's a lot of effort behind that post.

Heck, did you even read the post you replied to?

[Koub]

Actually, I'm neither talented enough to speedrun not smart enough to tell speedrunners in what aspect they might be wrong on their strategy. However, at my day job (which has nothing to do with Factorio), I'm considered an expert in my field, and I'm always amazed - and a bit annoyed - to see so many people who have no experience of my core business and who are convinced they know better than I do in my expertise field.

As for the reading the first post of this thread, I did read it, and even get back to re-reading it several times, because the more the thread goes, the less I understand in what it has anything to do with the OP. I'm just telling myself that as you (the OP) are very active on it, it goes where you wanted it to go, or you'd have complained by now.

If telling the speedrunners they are wrong on their expertise field was not the primary goal of this thread, what exactly are you tring to prove/teach/tell us all ? I'd like to stress that this is a genuine question, not trolling, or what, I really don't see the point of all this thread, and every time I re-read the OP, I'm confused, because there are tons of different subjects tackled, and I don't understand in what they make a coherent whole (the non understanding could be my imperfect mastery of English language though).

[jodokus31]

The main argument, which I understood is that bots are so game changing, that they are prio 1 for speedrunning any% and that Gotlap is almost right before bots in 25 min. That this is better, needs to be proven and i doubt it will work. But ok...
The second part is about deathworlds and how the are already a race implicitly, but could be more competitive. It's a bit off topic imo.

[blazespinnaker]

I'm considered an expert in my field, and I'm always amazed - and a bit annoyed - to see so many people who have no experience of my core business and who are convinced they know better than I do in my expertise field.

You're making the same erroneous assumption speedrunners make. That the only expertise that matters is purely about keyboarding.

It most absolutely is not! Yes, keyboarding is critical in the end, but the math of order flow and type of construction is most important.

However, keyboarders will always win out in the end, because everytime a new idea / order flow is introduced, they will copy it and execute very quickly, probably faster than the person who came up with the idea. But did they come up with the original idea around optimization? No, they just copied the idea, but executed it much faster.

Let me give a few examples: the miner feed into boilers, the double long inserters. Those were introduced a long time ago by some forgotten player (Phoenix came up with a lot I think or at least stole them cleverly, kudos to him).

But beyond the keyboarding, as far as I'm concerned, relative to them, I am the expert at construction flow. That said, there are definitely factorio folks who have come and gone which are I bow down before, and are still around (call out to Dave). And there are very likely lurkers around here sniggering at my naive attempts to LARP JIT scheduling expertise.

But the speed runners, by and large, have no clue as to what they are doing. Trust me, I've asked them what they are doing and what their alg for optimization is. Do you think any of them have read up on JIT optimization techniques and algorithms? Please. They are just doing it all by trial and error. The GOTLAP folks are close, but they should focus on bots and not trains.

However, let me be very clear - I would never deign to tell a speed runner the best way to lay out their short cut keys or how to stroke their fingers across the keyboard. I truly suck at that, no kidding. They are also much better at creating factories that fall to keyboarding easily. That is one important creative skill here that I don't focus on, though sometimes think about.

But they have very little understanding of what they are doing when it comes to over all optimization, that is very obvious to anyone who looks at what they do. They have no concept of the construction flow search space, the maxima/minima problems, and how to characterize them. I know this. I've take the the time to ask them all this question in discord chat.

This all said - I hope it's obvious though that I and others stand on the shoulders of giants. I steal ideas just as aggressively as anyone else (though I call out my victims as much as I can). But that's a very good thing, and something we should all do. And let folks LARP a bit of pride in it, is not a terrible thing. It's give and take. It's how we will eventually create the perfect game, which I think is a pretty fun community achievement.

And it is a community achievement, which will come from a mix of skills - math, keyboarding, creative factory building. It will not be any particular individual that will be responsible for it all, as much as they think they are the "WR" holder.

I should say that I don't mean to tarnish the runners too negatively. There are some really great folks, like Nefrums, who frankly has likely brought in massive revenue for Factorio via his soft people skills. His videos and speed running PDF and streaming are extremely popular and you can tell everyone is rooting for him. He inspires engagement in the game. I often wonder if these types of folks have some kind of rev sharing deal. They should!

[blazespinnaker]

The main argument, which I understood is that bots are so game changing, that they are prio 1 for speedrunning any% and that Gotlap is almost right before bots in 25 min. That this is better, needs to be proven and i doubt it will work. But ok...
The second part is about deathworlds and how the are already a race implicitly, but could be more competitive. It's a bit off topic imo.

Yeah, I threw the DW stuff in because people might find it easier to grasp. Nobody ever seems to discuss the bots thing intelligently. Hell, I might be wrong, however I don't buy the "but infra!" argument. That's not an argument, it's just a word.

If they can't see the bots exponentially large (relative to bot count, which can increase from bots themselves!) optimization search space and potential, I don't think we have a baseline set of facts to even disagree about.

Just In Time(JIT) is a real thing, and solving/optimizing JIT production chains is an extremely interesting and fun problem. With factorio, we can live action role play being JIT masters. And we get bots, too boot. For me, that's hella fun. But, I like to take these things to their conclusion. My wave defense post I think represents the potential wins from JIT extremely well:

viewtopic.php?f=5&t=90012

[blazespinnaker]

If you are interested in at least getting an over view of the problem, here's a recent paper to start with - https://www.researchgate.net/publicatio ... e/download

Obviously, unless you're an expert in this area or at least have some advanced familiarity with data structures and algorithms, much of it will be very obscure. Still, many of the words should be familiar to factorio players: logistics, supply and demand, operational efficiency, waste minimization, distribution of raw materials and finished goods.

From there, you'll want to look at the references and at least cycle through the abstracts. This will at the very least give you a sense of the problem space and how it's being tackled.

In theory, I have the technical skills to take it to those extremes, but certainly not the time or money. Rather, I'm trying a simpler approach similar to what was used for Starcraft optimization (GA), though I'm beginning to wonder if there might be an easier way which has more assumptions hard coded.

viewtopic.php?p=521638#p521638

I encourage you to look through the links I referenced in that post. GA algs are actually pretty easy to implement, but they get stuck in local minima/maxima too easily unless you hack them endlessly. That's because they are just randomly pairing parents. Deep learning optimization algs have similar problems, but more sophisticated techniques for getting past these max/mins. But deep learning has other problems and likely not easier to implement than GA. I've also found when trying to apply it to things other than image parts it can memorize, it hasn't been that effective.

[Koub]

[...]

Okaaaay ! Now I see better . I admit I was under the impression that to get so good timings, even "just" for an any% run, it was unavoidable to have done that sort of optimisation beforehand. Actually, if I had had a dream one day to be a Factorio speedrunner, I would probably have started by that kind of workflow study, starting from the end, and unravelling back to the beginning, to identify the potential bottlenecks along the critical path. And that, despite having no expertise on the workflow optimisation (neither academic, not tool based), just some kind of empiric approach with the most universal tool ever (Excel spreadsheets of course ), which should be better than nothing I guess. I assumed it was something natural. I acknowledge this wouldn't work for everything, but should for Factorio (to some degree).

This all said - I hope it's obvious though that I and others stand on the shoulders of giants. I steal ideas just as aggressively as anyone else (though I call out my victims as much as I can). But that's a very good thing, and something we should all do.

That's basically how mankind arrived to the point it is, by clever people continuously "stealing" ideas each other, using them for things they were not designed for, but work unexpectedly well, improving on them, ... So, yeah, I guess it makes sense

Have you tried to get in touch not with the whole speedrunning community, but just one speedrunner you might feel more receptive, and give him what would be a more optimized work order list based on your calculations (might take some time) ?

[blazespinnaker]

yeah, I had made the exact same assumption and had been very surprised when I kept on asking and bugging folks about what they did and they said they didn't really believe in that theoretical approach. I've done google searching on this and folks have suggested it, but nobody ever followed through afaik.

It seems like one of the areas of factorio that never really got a thorough treatment.

Actually, that is one of the main reasons I find it so fascinating - because it seems like a large, generic thing about factorio that is left to truly discover and it's about optimization, which I find very perversely very entertaining.

If anyone knows better and this has been seriously solved - please let me know. Would be very interested about how it was done and the algs used.

[blazespinnaker]

Have you tried to get in touch not with the whole speedrunning community, but just one speedrunner you might feel more receptive, and give him what would be a more optimized work order list based on your calculations (might take some time) ?

Yeah, I spent about two or three weeks on the steelaxe (factorio speedrunning) discord chat channel getting to know all the principals. I chatted with all of them, though sadly phoenix didn't join that often or at least wasn't very chatty. I brought this topic up a number of times but there was only dis interest, and not very polite dis interest Guys and video games, I tell ya. There were a couple of folks who were friendly about it, but other than showing an interest in results when I had some that was the limit of their discussion.