Subotai 0.9

[Mandulis]

Here comes the Subotai AI. Named after the famous mongolian general I will try to make it earn its name. It only plays mongols and 1vs1 matches at the moment.

[Campeador]

You've got only two errors:
- one of your timers lacks its constant
- your TSA file should be placed in a folder called Subotai, no in the AI folder

Apart from that, it's only a thing to complete and optimize your AI. I see that you are doing the same that I did when I started, to copy the Training AI and customize it to add your own strategy. In the future you'll change the structure to make it more "yours" and stronger. Your are going in the correct path.

Welcome to the world of .

[Promiskuitiv]

Hey Mandulis,

welcome to our lovely community!


I'll happily download and criticize the next version of your AI (with the bugs fixed).

[Mandulis]

I had the problem of getting endless monasteries built. I asked the guys in the chat and uploaded the script here to let them have a look. The script is horribly ripped apart as I deleted parts to find the error. The actual script looks a bit better.

I'll post the script, when im done with the first version. Maybe I find something smart to include.

[Campeador]

Well, in the version that you have uploaded here, it builds monasteries with a single rule:

Code: Select all

(defrule
	(building-type-count-total town-center > 1)
	(building-type-count monastery < 1)
	(can-build monastery)
=>
	(build monastery)
	(chat-local-to-self "building monastery")	
)

It haven't to be a problem if you play without the UP, because without it AIs can't buildmore more than a building of the same type at once. But with UP, using building-type-count (which only counts finished buildings) instead of building-type-count-total (which also considers buildings under construction) can cause more than a monastery at one. Until one fo them becomes finished. If in the version in your computer you have added something more then it's a different story.

[Mandulis]

The error was actually a different one. One of my deconst variables had a reserved name ("castle") which caused the AI to always build a monastery when it tried to build a castle. I already fixed that. The next task will be completing the castle age. There is quite some stuff to do. Although there a probably already well known ways which research to do in which order and so on, I want to find that out by myself to increase the learning effect. Let's see, how it turns out.

[jdstankosky]

22:11 scripter16 found the problem
22:12 scripter16 (defconst g-ranged-unit 24) ;which ranged unit to train primarily
22:12 scripter16 (defconst infantry-threat 24)
22:12 scripter16 you have 2 goals
22:12 scripter16 with the same internal id
22:12 scripter16 they are overwriting each other

[marathon]

(defconst primary-unit archer-line) Bad
(defconst primary-unit archer) Good

[jdstankosky]

(defconst primary-unit archer-line) Bad
(defconst primary-unit archer) Good

This is also my opinion, but scripter told him it'd be fine this way.

[Mandulis]

(defconst primary-unit archer-line) Bad
(defconst primary-unit archer) Good

Why's that so? I tried both and both worked fine syntax wise. But I am always ready to follow advice from the scripting veterans.

I'll upload the script again soon. Currently I am working on the basic structure. I want the script to have the right layers of decision making and to pursue it's own strategy for now. Adapting to the enemy and optimizin things (like spreading units and stuff) will come in step by step.

Maybe one of you could give me a hint if my plan is total rubbish or actually doable. My idea was to have a main script which handles everything that is not related to a specific strategy. It does the defconst stuff, the resign rules, taunting and trading. It also contains the decision making which is done in several layers. Layer one is scouting. Layer two is the analysis of the enemies possible strategy out of the scouting and the current circumstances (which age is everybody in and so on). The third layer defines my own strategy out of the enemy's strategy and the current situation. Here I will load the corresponding strategy script. Each strategy script will do the fourth layer. It decides on the primary melee and ranged unit I will build. The strategy script also contains the research and building rules as well as attack conditions.

Threat evaluation and training of units is put into the main script. Each unit will be trained out of three reasons only:
- it is my current primary unit
- it is the right unit for a certain threat which is currently active
- it is a subsitute unit (light cav when we want knights but are low on gold or so)

What I am trying to do here is to keep the training rules simple and short. Building and research rules should also be quite short. Of course the decision making has to be more complex to achieve this. I hope that the script will be quite modular in this way. Adding a new strategy would only involve the right detection and a separate script for building and research.


Probably I did only come to conclusions many already had before me, but you never know. So can this be viable or is it rather stupid?