User talk:Dsekercioglu/Thoughts on targeting

e.g. RaikoMicro is a direction based random movement. It choose whether to reverse direction or not on each tick, with a probability based on Bullet Flight Time.

Since the probability of reverse does is not affected by how long it has moved — TimeSince attributes will only be noise.

My gun proved that — the movement profile looks exactly the same on each segment of TimeSinceXXX.

Anyway, I found TimeSince attributes work surprisingly well against WaveSurfers. I've heard that PatternMatchers are good against WaveSurfers, maybe TimeSince attributes are one step to PatternMatching?

My scores had improved a lot back then when I boosted Time Since Decel's weight on the Easy part of TCRM. I thought it would be correlated because if you are changing directions every tick with the probability of x you would basically be changing directions around 1/x ticks. Let me check again... . The double Time Since Decel version did a little better(0.02) which is probably just noise. I am going to look at the profile... I am not getting flat profiles here. When I try 11 different time since deceleration from 0 to 1, values of the Guess Factors are like:

45
45
45
45
45
45
30
30
42
21
42

After 150th round I started to get values similar to this one which means it makes a difference for WhiteFang. It might be the 1 / (1 + 0.1 * x) formula I am using.

Edit: To make a fair comparison I tried this with both all weights 1 and my normal weights.