[3.2+] NullBot 3 ‘Reloaded’

[CDR Manuel]

Also, I wish that I could have NullBot fight against NullBot on a 2-player map.

[Legendary]

Does the NullBot v3 is still in development? On Github I see last commit a year ago...

[NoQ]

Not much; i had a couple of plans for making it play better, but it mostly works. I might return to it when i have some time. We also planned to replace nullbot with nullbot3 in master, a bit closer to some kind of release.

[Legendary]

I keep hope in you NoQ NullBot is a real challenge for me and countless hours for WZ matches. Thank you NoQ! Maybe you have any donation "button"?

[NoQ]

No, thanks, doesn't quite work that way Thanks anyway (:

[montetank]

I keep hope in you NoQ NullBot is a real challenge for me and countless hours for WZ matches. Thank you NoQ! Maybe you have any donation "button"?

Did you ever try to play against a team of 2 Nullbots in a 3 player map? There are many options and permutations possible. Let one Nullbot research the rocket tree, the other Nullbot the cannons a.s.o

[NoQ]

Here's the shape in which i'm planning to include nullbot in one of the upcoming betas.

• nullbot-3-04.wz

This release includes the remake of the hover personality (available in nullbot2 and included in warzone2100 v3.1) on the nullbot3 engine. Perhaps the personality needs a better name, because it's become clever enough not to use hovers on land maps and on pure-air maps, so it might break your expectations if you expect the AI to be as stupid as the caption says (: anyway, hope it works.

And i tweaked the early game behavior a little bit to adjust to the balance reverts. Now it should start much more sensibly/aggressively; machinegun-based personalities should rush ahead with mg viper wheels, flame-based personalities would tech faster in early game, and the turtle AI should now grab more land and try to set up more defenses, without worries of running out of power.

[MIH-XTC]

Hey NoQ,

It seems that Nullbot in 3.2 doesn't research as continuous as it did in 3.1.

I'm playing a 5v5 map, each player with 9 research centers to simulate a flawless research game but the AI doesn't continuously research in 3.2 as it did in 3.1. There are long periods of time where the AI could potentially be researching an item but isn't.

I tried on medium and hard settings but it made no difference. How frequent or on what trigger does the AI make a call() for research item? I think AI should have flawless research, at least for now. I thought it was that way in 3.1 but I'm not sure.

[NoQ]

Do they have money? They don't click the research now unless they have positive money including already paid items.

[MIH-XTC]

Do they have money? They don't click the research now unless they have positive money including already paid items.

Ah, yep that is it. They don't have enough power.

hmmm, that has its pros and cons. I don't know if that's desirable or not since there is the chance of research going stale.

Anyways, this solves my problem of simulating the tech tree. Thanks.

EDIT: maybe not, seems to happen on T1 half base but not on full???

[NoQ]

Once the AI has money, it tries to spend it on research first, and on other things only if spending on research fails. And i think it prioritizes research a lot more strictly than when the old AI had a bunch of non-research stuff waiting in queue - even though it produces less hearts in the tech list. Additionally, the AI no longer utilizes all labs when he has too little income, because he'd be too much outnumbered otherwise.

EDIT: maybe not, seems to happen on T1 half base but not on full???

Hmm, that'd be weird. Will check.

[MIH-XTC]

Hey, I found that the research problem I was describing earlier was inherent to the map I was using so apologies for that.


Question though, since I'm just getting into AI development. What's the reasoning for using javascript as opposed to the .vlo and .slo files? Perhaps a better question is; will it be ok to continue to use the .slo and .vlo files in 3.2 and going forward? I just spent a considerable amount of time fixing Nexus AI only to realize the other AI's are using .js so I want to know which route to go. I'm sure these questions have been asked somewhere already I just don't know where in the forums. Thanks.

[Berserk Cyborg]

Question though, since I'm just getting into AI development. What's the reasoning for using javascript as opposed to the .vlo and .slo files? Perhaps a better question is; will it be ok to continue to use the .slo and .vlo files in 3.2 and going forward? I just spent a considerable amount of time fixing Nexus AI only to realize the other AI's are using .js so I want to know which route to go. I'm sure these questions have been asked somewhere already I just don't know where in the forums. Thanks.

wzScript, as it is called, is an old API and really not worth the effort to create anything with anymore. It can and will over complicate every situation and is harder to track down bugs in. Javascript can achieve what you want with much less code and makes the code much easier to maintain and improve upon. So I highly recommend using javascript over wzScript.

Nexus and Semperfi are both written in wzScript (as well as the majority of the campaign). Not sure where else it is used.
Edit: Typo.

[Per]

Also, all future improvements will be added to the javascript one only. Since both scripting languages use mostly the same trigger-based way of running, it should not be very hard to port your code from one to the other. I think you will be pleasantly surprised at how much easier it is to work with the javascript.

[MIH-XTC]

Okay, thanks guys. I'm aware of the Javascript conversion initiative and have been reading the jscam thread but that thread hasn't really made sense to me until just now because I haven't seen any scripts in WZ yet, just the c++ source. I wasn't really sure what was being converted or what wzScript was. It makes sense now seeing the javascript and wzScript usage in AI. I'll take a look at the .js AI's more next weekend.