159 replies to this topic

[Pureon]

Quick video of Marilyn in action:



I didn't garrison once throughout the match

Here is the log: http://pastebin.com/E9111zEi

[wraitii]

Ouch, that hurt . i've got to learn to play this game properly...
Anyway, thanks for the test... I noticed Marilyn being quite confused when you attacked during her attack, which shouldn't be... Will look into that and this error.

[Pureon]

No problem, hope it helps

Another thing to avoid:

[wraitii]

Yeah, Mythos and Bruno seem to be on that... I've sent him a PM about this, but then I figured that it's probably not too expensive to count the number of workers on any given resource at any time, so the PM I sent him is really a non-issue... I intend to fix it, and along with some clever UnitAI, it'll be.

[wraitii]

Allright, following Pueron's advice, I'm revising the economic manager, and making interesting progress. The new idea is that I keep a list of resources that I want to collect from for each resource type (ultimately, perhaps subtype too), and then I'll assign the most efficient workers to each. Still much of a WIP, but I think it can be very efficient.

And as it seems like Quantumstate added a way to count the number of entities working on a resource at any given time, I'll probably get some inspiration to do that from him.

[wraitii]

Allright, I originally intended to implement a complicated resource gathering system... Only, it's too complicated to be any kind of practical (because I need to plan for future unit creation, mainly).
So, I'm changing my plans to a much simpler system: I'll assign my females to food gathering, and citizen soldiers to the rest. This won't change during the game, unless there is a particular need (fast collecting for a fortress) or a discrepancy (say, after an attack). It should prove efficient, as it's what most players do, with great success.

[Pureon]

That should work just fine. Looking forward to it

[wraitii]

It works fairly well as long as the training follows... The hardest part is dealing with sudden needs (such as a fortress build) or with the aftermath of an attack, because you might want your females to chop wood or stuffs.
Testing a very basic version in the early game against the latest qBot, it seems about 600/700 resources more efficient...
Anyway, don't expect it too soon, I'm going to have a very busy week.

[wraitii]

Allright, so being all busy with my work, and then wanting to rest, I haven't touched this in a few weeks... But I'll probably resume working on Marilyn soon. I've been able to at last actually play the game since I had a friend at my school download the game and we did a little LAN. Motivated me quite a bit. It was the longest time since I last played against a human (AOE II, how you remind me…).

Edited by wraitii, 23 June 2012 - 08:17 AM.

[wraitii]

Looks like I've lost some of the last progression I had done... Not too bad, since I was stuck in a bad place. I'll recode in the simplest way I can a way to have females gathering food and the other wood. Now on the other hand, I'll try to make it even more "self-aware" because while the architecture is in place, it's not really used... And increasingly prepared to technologies. Using my newly found knowledge of the game, which I acquired by playing it.

[Pureon]

Good to hear

[wraitii]

Allright so I'm getting fairly good results in the early phase of the game (beginning ressource collecting) for now. I still have a ton of things to keep doing, but it's usually on par with latest qBot or better. I'll have to compare with myself.
Oen thing I'm trying to do, for example, is having the flow of houses keep with the flow of citizens, and having Marilyn build the best citizen soldiers (for most civ, that will be the fastest unit costing wood/food. See for example the difference between the peltast and the foot companion for the macedonians).

[Arthur_D]

What I personally would like to see is an AI that doesn't stop doing stuff and starts derping for the rest of the game after 45 mins. or so, which is currently (latest alpha) often the case with both qBot and JuBot. Also, an AI which actively targets incoming rams would be great. Right now I can build 10 rams, and use 30-40 soldiers as cannon fodder, and while they are mucking about I use rams to take out civ centres and castles. So if the AI actually tried getting rid of rams as fast as possible, that whole operation would be much harder, and make the game last longer. And if the AI copes well with longer games, that won't make it a drag either.

Edited by Arthur_D, 28 June 2012 - 02:03 PM.

[wraitii]

The problem with qBot and Jubot, right now, is that they use the same strategy throughout the whole game (with minor adjustments for the early game). I'm implementing Marilyn right now to change its behavior as time goes by, which should allow for a better control of the late game.
Attack and defense also need to be improved a whole lot, of course. However, for now, I'm trying to get the economic system to be resilient and efficient. I've got the "efficient" part done, now I must teach it to react if it's attacked or something. And some other economic stuffs here and there. Then I'll focus on other stuff.

Edited by wraitii, 28 June 2012 - 02:21 PM.

[Arthur_D]

Sounds good, keep up the great work.

[wraitii]

I had centralized a lot, everything in "Headquarters"... I'm going to decentralize a tad, or it'll get messy. I'm currently working on proper resource management, as I will need that for my economic manager to work (by that, I mean: I just started. Was busy porting fancy shader-y stuff.)

I'm kind of backtracking from my "UnitInfoManager" and "BuildingInfoManager" idea... Storing in entity collection was the right idea, but my implementation is not always practical. These will probably remain in use to decide which unit/building to train/build, what is available, these sort of things.

Edited by wraitii, 09 July 2012 - 04:49 PM.

[wraitii]

Bad news. I'm pretty sure, after much thinking, that I'll never have time to recreate the economic manager completely and properly. Thus, I'm abandoning Marilyn's development to focus on merging some of it into qBot. 1 working bot and one sub-par experimental bot are not interesting compared to having one really good bot, I believe.

Thanks to those who followed the development, tested the bot and motivated me . And expect qBot to feel a little bit blonder and more feminine over time.

[Almin]

I just wanted to thank you very much for your work , wraitii! I looked over your code to learn form it and to understand how it works and it really spent me some experience. Thanks for sharing it. Hopefully your next projects will succed.

btw: Today is the 50st anniversary of the surprising death of Marilyn. (Isn't that strange? )

Edited by Almin, 02 August 2012 - 05:26 PM.

[Geek377]

wraitii, on 02 August 2012 - 09:54 AM, said:

And expect qBot to feel a little bit blonder and more feminine over time.

No problem!

[rjs23]

Keeping my eyes on this thread for progress!