entering

Jump to navigation Jump to search

You do not have permission to edit this page, for the following reasons:

  • The action you have requested is limited to users in the group: Users.
  • You must confirm your email address before editing pages. Please set and validate your email address through your user preferences.

You can view and copy the source of this page.

Return to Thread:Talk:RoboRumble/Enter The Competition/entering/reply (36).

Thank you very much. Thanks for your help.

Dsekercioglu (talk)19:49, 3 September 2016
 


My bot WhiteFang uses auto-segmentation. Is it a good tactic and are there other robots doing that?

Dsekercioglu (talk)16:59, 4 December 2016
 

I mean that it tries to predict the best segments that could be used against it's enemy.

Dsekercioglu (talk)17:26, 4 December 2016

Most top bots simply use all segments at the same time, but with different weights. They are usually tuned offline.

There are some top bots using 2 sets of segments. 1 for a general purpose gun and 1 for an anti-surfer gun. And the best set of segments is estimated on-the-fly.

MN (talk)05:08, 5 December 2016

MN thank you. I used all the datas the average hit rate increased like % 0.8. After I added auto-tuning for setting weights and average hit rate increased about %0.9.

Dsekercioglu (talk)20:20, 5 December 2016

The 2 segments which differ most between different opponents are the real/virtual wave segment and the data decay (time) segment. These are the 2 usually separated between "general purpose" and "anti-surfer" sets of segments.

Note that using many different targeting strategies for different situations may slow down learning speed (time wasted auto-tuning the gun for the optimal strategy). Because of this, most top bots use only 1 or 2 guns.

MN (talk)22:30, 5 December 2016

I don't create a new gun and start it from the beginning. I test the segments alone and when they hit I multiply it with a default. So it keeps learning and uses the same kd-tree all the time. Also according to my results it increases learning.

This is the code of Tuning:(There are more code for intersection but the most simple one is this)

           if (Tools.getDistance(currentX, currentY, myXD, myYD) < 72) {
               AntiBotTuner.movementScores[index] *= 1.1;
           }

Usage of weights:

double diff = Math.abs(p1[index] - p2[index]) * dsekercioglu.AntiBotTuner.movementScores[i];

In the future versions I will also use an Anti-Surfer Gun.

Dsekercioglu (talk)15:26, 6 December 2016