Tournament Analysis - Why is RalphsAI losing to Convoy?

[Zutty]

The tournament system is fantastic! What a brilliant idea. The scope for analysis alone makes it worth its weight in gold.

While browsing the overviews, one matchup between Convoy and RalphsAI caught my eye in particular...

http://devs.openttd.org/~noai/tournamen ... erview.gif

The levels of income, profit, and performance are all very competitive and both the road networks look good. They're serving the same places in some areas and there are is obvious wastage (I haven't actually watched it for myself yet in game).

So why then is Convoy's company value so much higher? If Convoy has more assets (buses, stops, etc...) then how is RalphsAI able to compete? By 1960 they're level pegging on income.

Ralph - I really REALLY don't mean to offend you at all in this post. You AI is clearly very good, it just seems to be missing something, though I'm not sure what.

Thoughts?

[Ralph]

No offence taken , I had played mine v Convoy a few times and was not expecting it to do so badly, but I used the operating profit graph rather than company value as a metric.

I think the main killer of company value is the loan, Convoy typically pays it off in about 5 year. I don't bother, as even after a couple of years the interest is not worth worrying about.

Also Convoy has a better start as it does not reuse towns until it has to, the first thing my AI does is connect one town to its 3 nearest neighbours, and while maybe slightly more 'human', its quite inefficient as you have a much smaller pool of potential passengers, this is on my list of things to change as I did it quite quickly the first time.

Interestingly mine does much better with a slightly longer game (15 years is fine) because it replaces old vehicles, which Convoy does not do yet.

[TrueBrain]

One might wonder what is better: a human like AI (I personally vote yes for this), or a tournament capable AI. Although the tournament system is nice, having an AI that acts more like a human, and doesn't completely overrun any possible human player, is nice too But I guess an AI should be configurable to be either one

[Finaldeath]

I think the AI which a player enjoys is more important, but I, and I am sure others are, going to put in a ton of parameters so, if the player wished (or for tournament play) you could set it up to be the "most competitive".

One thing is that there is no optimal AI, so tournaments might always be close - and noting of course that some vehicles have short-term profit and others longer term profit. The former is things like early coal, the latter might be ship routes (with 30 year life spans for the vehicles, it takes a long time to get back money) and bus routes combined with airports when cities have grown after 10 years of buses.

Could have the two AI's work on different goals, and both come out on top with competitive play - certainly if they both do different things and don't actually compete with one another!

[fabca2]

One might wonder what is better: a human like AI (I personally vote yes for this), or a tournament capable AI. Although the tournament system is nice, having an AI that acts more like a human, and doesn't completely overrun any possible human player, is nice too But I guess an AI should be configurable to be either one

You totally right, and I agree with your vote.
but it will mostly depends on you (as being main dev).
- will human player will be able to choose (mark/unmark) which AI are allowed in the game ? (I know it's too early to ask such question)
- which genuine AI will you ship with ottd ? (I mean after the merge with the branch, will you ship only the contest winner ?)

what you made is really really great, and you will have to think how can it be profitable as best as possible for the gameplay/users.
Thank you.

[TrueBrain]

You totally right, and I agree with your vote.
but it will mostly depends on you (as being main dev).
- will human player will be able to choose (mark/unmark) which AI are allowed in the game ? (I know it's too early to ask such question)

The player will be able to select which AI he wants in his game. In this GUI it will also be possible to configure the AI, if the AI allows that.

- which genuine AI will you ship with ottd ? (I mean after the merge with the branch, will you ship only the contest winner ?)

This indeed will be a very hard selection process Even more as we require certain standard of such AIs (coding-style). I am considering building a small website where people can upload their AI, and give a description and screenshots and stuff, so people can choice for their self which one they want to play against. This of course will include contest winners, if they allow their AI to be published of course

what you made is really really great, and you will have to think how can it be profitable as best as possible for the gameplay/users.
Thank you.

No, thank you