Talk:Super-Melee tactics

From Ultronomicon
Revision as of 13:43, 19 December 2007 by Zeracles (talk | contribs) (Deconstructing the UQM Source?)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Perhaps a good idea for this page would be to put a grid with all the races across the top, and again down the left side. Then, you would have a square for each possible match-up, such as "Ur-Quan Dreadnaught vs. Druuge Mauler". users could click on some clickable thingy in the box, and it would jump down to a # spot on the page. or, of course, they could just scroll down and eventually see it. That's all, of course, if you want to have strategies that break down to every possible combo.

Hey, you're Shiver, right? Anyway, what you say about a clickable grid sounds good to me. I'm not sure I understand what what you mean with the last sentence though. I don't think it would be necessary to break it down that much, remember that we have pages for generic tactics like pillboxing which readers can look up and we can put links to so we don't have to explain them down to the last detail every time. And we could define more of them and create pages as necessary. Cheers. --Zeracles 04:28, 16 December 2007 (CET)

Strategy Space

Okay, so you're not Shiver, sorry, this is something of a coincidence. Maybe you only read the thread . . . anyway, we should show him that it is the destiny of any private project to be superceded by open, collaborative efforts (a la unix, wiki, . . .) heh heh. I have an idea for this page which would involve a massive amount of work - I'd like to model all the strategies via the code, seeing what happens empirically. I wouldn't get onto it until finishing with the planet lander formula (and yes, I believe we will finish this at some point), but there it is.

Now, before you say "but that's ridiculous, such a huge amount of work to probably only tell us what we already knew", I think there's a lot we could learn from modeling like this. Which I will elaborate on soon. --Zeracles 12:04, 17 December 2007 (CET)

Also, Zeracles: There has been a thread on the UQM forums about improving the Super Melee AI. Using a learning AI was proposed, so maybe the posters in that thread could help you. Valaggar 13:22, 17 December 2007 (CET)
This is interesting, thanks Val (now if we had a finished SCFU history already I might have known this . . .). But these people are talking about a learning AI, when I say see what happens empirically, I mean something more along the lines of the iterative approach I've proposed for the planet lander formula. I don't know how much attention you've paid to this, Val (I don't mean anything by this, to an onlooker it probably has lost a lot of the immediate interest), but what I do here is try every possible strategy, and just see what works. In my view this is the easiest way to find a best strategy. At the end of such a process, we will be able to say, ``look, we tried every strategy, this was the best". There is nothing contrived about such a process, and so we would pass the test of objectivity.
This may not be appropriate for the content of this page, of course, I'm just highlighting a possible direction. Also, the learning AI approach is probably not so different from this iterative approach. Both these ways are about learning from experience, are they not? I'll come back to the question of looking for help soon.
As for how we would do this exactly, if it was me I would try to do it the same way I've started to do it on the planet lander formula. We specify each strategy by a set of numbers, a vector. So every time we try a strategy, this just involves changing the numbers. Now, I can see comments in this thread which suggest that the code is somewhat recalcitrant (meaning no disrespect to the Twin Gods), so I'm guessing we would have to rewrite the code for it to take these vector-valued strategies.
For the Eluder (my favourite ship), say, a rough strategy might be specified by the range at which one turns and lobs BUTT missiles (a number) along with battery level threshold at which one comes back to launch another salvo (another number). So here's a vector of length 2, which specifies a range of strategies. Write the code to accept these, and away we go, which works best? - against each strategy of each other ship. And here's where we hit turbulence. For each of the few hundred possible match-ups, one would have to try each strategy for ship A versus each other strategy for ship B. Not to mention, it would not be trivial to assign vectors to the full range of melee strategies. At least the sa-matra only has one strategy . . . or does it? --Zeracles 14:09, 19 December 2007 (CET)

Deconstructing the UQM Source

It seems to me that many have tried their hand at working with the code without leaving behind good notes on the code's structure and what things they learned about the best ways of achieving best results. I think that such a thing could be very useful for any who are interested in this sort of thing (just about everyone). And so if I go through with any such work, I think it would be good to leave behind a pedagogical deconstruction of the code and what makes it tick. I mean a kind of ``for dummies" guide or tutorial. I assume that such a thing doesn't already exist.

So maybe I'll start up a user page to this end at some point, or we could put it in the main body of pages if we think it appropriate (do we?). It would be a preliminary project. Lastly, I should say that I know next to nothing about C. As much as I try to avoid it, learning languages on the fly is part of research though. --Zeracles 14:43, 19 December 2007 (CET)