Relationship between TCRM and hitrate
And from practical perspective, my answer is no again. I have different guns with different tcrm score, however the best gun in tcrm generally performs even worse in rumble
The difference in tcrm score of those guns are ~0.5, sounds quite a lot, however after calc I know it’s only 0.001 However the best gun in rumble could score ~1.0 more APS than the worst (which, ironically, is the best gun in tcrm score) So I think it isn’t wise to improve tcrm score any more once the result is already among the best.
Maybe for gun challenges, we should create something more relevant to rumble score then.
I would vote for the targets to fire a 1 power bullet away from you each time you fire. Then it would be more sensitive to small changes even at higher levels.
I am not sure about that. You get a bit more difference in bulletscore, but less difference in survival, at least against the good movers (which are the most interesting)
Well it will affect both. Basically then we actually have a direct measure of the hitrate, and not just a measure of the amount by which we run out of energy.
The score wouldn't go to 100 anymore though so we would have to scale it.
I think this is better than a non-firing target, as hit-rate is not the only thing that affects score.
If you choose to fire only when a guaranteed hit is possible, hit-rate will be 100%, however you merely earn score this way.
Then a target that losses energy which is the score you can earn potentially could better simulate real world situations
I think the reverse of mc2k7 could be used
The reference bots will be using RaikoGun.java from mc2k7 for radar, gun and energy management
And the challengers use their own gun which fires 3.0 bullets where possible, and the movement should be a modified version of RaikoMicro’s, which allows prediction of one’s own movement. The appearance of the movement should be kept the same though.
Although bullet damage doesn’t scale to 100 in this case, since the only variable is your gun, all reference bots are given the same weight actually, scaling isn’t necessary.
I like the idea of the reference bots using RaikoGun.java and the challengers using our own guns.
What type of bots should we have in the challenge? Maybe we should make the reference bots kind of match the LiteRumble population, so maybe ~20% wave surfers, ~50 random movement, and the rest oscillator/pattern movements?
I'm rather clueless on the average amount of each archetype in the LiteRumble, but I think these stats sound about right.
I think we could separate each type of movements, and weight them accordingly instead. A patch to roborunner may be done though.
Anyway IMO it doesn’t matter how you can hit sittingducks and other movements, as long as you can dodge their bullets well, so the weight for pattern movement depends on your movement in real rumble, and should be much lower in general, despite the popularity.
another option would be to separate random movement and surfers into two challenges, since the strategy used to hit them generally differs a lot, it will be more convenient to make them into separate tc as well. So that gun development could be done without messing things up with virtual guns.
An overall ranking could be created though, by weighting the score of the two, given the challenger is kept exactly the same. This gives virtual guns a chance.
The problem of letting the reference bots to use real guns is that it further obscures the relationship between hitrate and score, and the randomness added from reference bots hitting you means more seasons must be used for reliable score.
However this makes bullet shadow possible, so that you can finally test your guns’ performance against bullet shadowers in tc.
Another option would be to use a snapshot of the entire rumble However in this case we can’t control the guns
So that each opponent maybe given different weights, where normalization should be used to compensate
I think this approach should be the most relevant one to rumble score, however it makes relationship between hitrate and score unclear, if different opponents aren’t normalized well