A.I. too overpowered

@gunshin would be a good person to ask but once you have a stable eco you can have one factory producing engies nonstop till the end of time.

I'd say one factory dedicated to build fabbers on a medium sized planet.
You might manage with less fabber production early if you don't expect to lose many fabbers and especially if you don't use your commander offensively and need a lot of fabbers to repair your commander.

I don't tend to build the dedicated fab factory until later on, but I usually make the first 1 or 5 units any new factory builds be fabbers.
As for the Ai, an expansion speed slider would be cool, the higher it is the more fabbers it builds and the faster it expands.

I think rather than "how many" the question should be "what percentage".

Either X percentage of every unit built be fabbers, or IF (Fabbers) Less than (Percentage of all existing units) THEN build more fabbers.

Probably less linear than that, you need fabbers before you need tanks, but you don't need many fabbers until later on in the game.

One factory non-stop building engineers, send out in groups of two or maybe 3, non stop expanding for mex, building AA as they go and also laser turrets next to any closely clumped groups of mex.

Would also be nice to build 2-6 bot/veh factories in mini proxy bases around the map.

Also don't spam walls around the pelters, they are insanely expensive and would be better spent building more pelters instead. (Not sure if they still do this)

Also because this would be beyond hilarious (and infuriating), program a group of engineers to lay a couple laser turrets and then pelter creep when close to the opponents base.

OMG THIS teach the AI to firebase. I'd love it.

BTW are such behaviours something that you have to tell the AI to do or is it something that the AI finds out on its' own?

Edit: I think it actually might be worth it to surround a pelter with walls if you expect large swarms of spam charging at you.
I would try it if the gentlemens didn't disallow it.

Short answer

There's no such thing as too many, have a factory doing nothing but building fabbers. Bot fabbers.

Long answer

As of current balance the first factory has to be bots and then you almost certainly have to start with about six fabbers, two for grabbing metal in one direction (with AA turrets for every MEX pair, or every MEX if you're really worried about raiding), two for grabbing metal in another (same with the AA), and two for energy and other general tasks. Sometimes you send three fabbers for expansion rather than just two. You never send fabbers to perform tasks alone and you don't use vehicle or air fabbers unless you want their tier 2 building because they're worse in every way.

This certainly isn't absolute, different players are trying different things, but really, you can't have too many fabbers, players use extra ones in the base just to quickly pump out their late game stuff, be it spamming factories or assisting rapid nuke building. The AI may be able to be a little fancier about this since a dedicated factory is to allow a human to turn their attention elsewhere and not have to worry about replacing lost fabbers or rejigging queues to do so, but the AI probably doesn't need to worry about queue management so much.

I think this is a pretty good example of your typical 1v1 and the kind of fabber usage seen right now



I count, what, eight fabbers for both players at the six minute mark? Also note the factory levels, one player has four with a fifth on the way while the other has three with their fourth on the way. I've never seen the AI go above four tier 1 and 1 tier 2, but really it should have tier 1 factories everywhere. Factories in the base, factories in clumps around the map to both allow attacks from unexpected directions AND prevent the enemy destroying everything in one go. For tier 2 I think Zaphod calculated that it's more efficient to have up to twenty two fabbers assisting one factory than to build a second.

After the initial eco build up players tend to drop down factories four to six at a time.

NO ai is a misleading term.

"computer opponent" would be a more appriate term
PA would be 100GB if there was a learn-capable algorythm inside.

You should check out Sorians work on neural networks in this lecture starting at 17.40:
http://www.gdcvault.com/play/1015667/Off-the-Beaten-Path-Non
It is learning.

Edit:I guess the alternative to calling it learning would be evolving or testing variables until it have better fitness and becomes better at playing the game.

There are many problems with generating genetic algorithms for such things. One of the things that are needed (well, best to have) in our case is an opponent who does exactly the same thing everytime, and have the algorithm attempt various ways of winning the game in better (how you define 'better' is another topic for discussion) ways. A genetic algorithm is likely as close to an AI we are ever going to get without 'true AI' s. It is also very hard to do. So i hope that if sorian ever manages to do it with PA that he actually writes some research papers or books on the topic ^_^

It would be an amazing experiment, but would require thousands of games testing on a single iteration of the algorithm in our case. It actually makes me want him to do it, but i also would rather him focus on the actual game itself as we are nearing release date and not all the features are in yet =O (heres hoping to him proving me wrong and doing both!)

Early access just paid for itself.

Hopefully other players like @brianpurkiss and @kingjohnvi will chip in with their experience, as I'm more of an outsider looking in when it comes to top level play in the beta.

Oh yeah, another suggestion, the AI should put it's energy in places that are easier to defend. I tend to wade through it's power gens on the way to the factories and gut it's economy.

Most skilled human players however seem to put factories between the enemy and there power gens, with the idea that the enemy will have to make it past the units to get to the power, and that factories are easier to replace than generators (won't crash the economy)

An algorithm that determines the point least likely to be attacked by the enemy or deepest inside the base, and puts generators there, would be cool.

i hope it doesnt get too hard coded, im kinda hoping the initial spawn bo will go from the obligatory mex - pgen - mex - mex - factory to factory - mex/pgen - mex - mex -mex (or similar. factory first would be a huge improvement).

I meant more around whether they just pump out fabbers or have goals in mind around the numbers of fabbers to minutes played, etc.

wow! really tanks for the linky! I'm going to have a blast watching this.

still though. the obvious with this is that the whole "learning" work for the AI will be on a workhorse machine. In all likelyhood not the same server we play on, hence we'll never get to witness "learning".

and also an AI of a magnitude of intelect, capble of working such things out as using walls around PDs would be a monster of an AI and take years of trial to start implementing this behavior. the easy option is just to tell the computer opponent to just place walls around PDs no "because", remember it doesn't think.
I may be spouting bs but that's what I know about AIs