|
|
| Line 1: |
Line 1: |
| − | July 6, 2019, Basilite 0.1: Ranked 5 in MicroRumble, 13 in MiniRumble, 99 in RoboRumble
| + | For several times I see a learning gun bumping from one direction to another direction, |
| | + | and yet each time an aim isn't satisfied, so no bullet is fired at all for a long time. |
| | | | |
| − | Features Basilisk’s movement and BlackWidow’s PM gun
| + | Actually this scenario disables the gun, until it gradually learns the new movement of the opponent. |
| | | | |
| − | https://drive.google.com/uc?export=download&id=1ksbMq8V2AHUbH3BwImEjH_kxl_7vJHAZ
| + | However this behavior doesn't make sense. A prediction should be consistent, |
| | + | at least for a short period of time (and until new information that denies previous assumptions appears), |
| | + | or you're basically saying your own prediction is wrong, by changing mind frequently. |
| | | | |
| − | ----
| + | So, analog to human brains, could we just introduce an inertia to gun prediction process, |
| | + | making it stick to previous predictions a little bit more, and as a side effect reducing noise? |
| | | | |
| − | July 7, 2019, Basilite 0.2: Pulled, had a bad bug
| + | E.g., introducing some rolling average, accumulating previous predictions with the newest ones. |
| | | | |
| − | Added a random offset to the gun to throw off bullet shielders
| + | This could be done at firing angle, but can also be done better, by doing the "accumulation" |
| − | | + | (merging previous cluster with newest cluster, and weight them properly, or adding things up with VCS bins) |
| − | If Math.random() > 0.5, it aims slightly to the left. Otherwise, it aims slightly to the right
| + | before the kernel density estimation. |
| − | | |
| − | Small tweaks to pattern matcher
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1hrKho4mYpj10LCH2l1RDD5kTCl0SzKwk
| |
| − | | |
| − | ----
| |
| − | July 7, 2019, Basilite 0.3: Ranked 5 in MicroRumble, 12 in MiniRumble, 92 in RoboRumble
| |
| − | | |
| − | Revert to 0.1
| |
| − | | |
| − | Add small random factor to gun
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1nxI2-XWMt70zsKidnJ5yy2KLNKy6VzI_
| |
| − | | |
| − | ----
| |
| − | | |
| − | July 7, 2019, Basilite 1.0: Pulled, turns out unsegmented guessfactor guns aren’t very good :P
| |
| − | | |
| − | Took out pattern matching gun for a simple, unsegmented guessfactor gun
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1bORs6OM5j4pIshREYUfd9W99hge_6gVk
| |
| − | | |
| − | ----
| |
| − | | |
| − | July 8, 2019, Basilite 0.4: Ranked 6 in MicroRumble, 15 in RoboRumble, 104 in RoboRumble
| |
| − | | |
| − | The PM gun is now aware of walls and tries to avoid directly firing into them
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1cQlPOZ8-5FDKwfb0qPl3MC1CUqfOTqMq
| |
| − | | |
| − | ----
| |
| − | | |
| − | July 10, 2019, Basilite 0.5: Ranked 5 in MicroRumble, 12 in MiniRumble, 95 in RoboRumble
| |
| − | | |
| − | 84.62 APS in MicroRumble
| |
| − | | |
| − | Tweaks to dive protection and mode selection
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1oUjXdg0HQ5kfXCAMKOerYGnzr6jVLc0c
| |
| − | | |
| − | ----
| |
| − | | |
| − | July 12, 2019, Basilite 0.6: Ranked 5 in MicroRumble, 15 in MiniRumble, undetermined in RoboRumble
| |
| − | | |
| − | 84.28 APS in MicroRumble
| |
| − | | |
| − | Reduced match length to 15
| |
| − | | |
| − | Tweaks to enemy log
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1_AIDMKnVmOvsPsncYnjQmXEleJlEPif4
| |
| − | | |
| − | ----
| |
| − | | |
| − | July 13, 2019: Basilite 0.7: Ranked 5 in MicroRumble, 13 in MiniRumble, undetermined in RoboRumble
| |
| − | | |
| − | 84.42 APS in MicroRumble
| |
| − | | |
| − | Increased match length to 40
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1F00xV-QivNFjY4zqwsySYMOn8tR-Fjyf
| |
| − | | |
| − | ----
| |
| − | | |
| − | July 23, 2019, Basilite 0.8: Ranked 5 in MicroRumble, 13 in MiniRumble, 94 in RoboRumble
| |
| − | | |
| − | Revert to 0.5
| |
| − | | |
| − | Decrement match length by 2 instead of 1
| |
| − | | |
| − | 84.52 APS in MicroRumble
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=10CI8vrHatNiCsR2ADFQyzYrL0WtBlue9
| |
| − | | |
| − | ----
| |
| − | | |
| − | July 20, 2019, Basilite 1.1: Ranked 6 in MicroRumble, 18 in MiniRumble, 161 in RoboRumble
| |
| − | | |
| − | 82.63 APS in MicroRumble
| |
| − | | |
| − | Now uses a guessfactor gun segmented on distance and velocity
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=106K42VlWHLIZxajvaOaqLh6wfHYNpRVy
| |
| − | | |
| − | ----
| |
| − | | |
| − | July 27, 2019, Basilite 0.9: Ranked 6 in MicroRumble, 16 in MiniRumble, 104 in RoboRumble
| |
| − | | |
| − | Match length increased to 60
| |
| − | | |
| − | 83.69 APS in MicroRumble
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1tRznmCcccugu7HQtueCdb23nRm6Wdnz6
| |
| − | | |
| − | ----
| |
| − | | |
| − | August 11, 2019, Basilite 0.10:
| |
| − | | |
| − | Improvements to random movement flattener.
| |
| − | | |
| − | Now remembers if it would have changed direction if not for the flattener, and changes direction later to account for it.
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1KZR0grxQtH_4RopIY9ggq5J-Fi_9r18q
| |
| − | | |
| − | ----
| |
| − | | |
| − | December 2, 2019, Basilite 0.11: Ranked 4 in MicroRumble, 12 in MiniRumble, 98 in RoboRumble
| |
| − | | |
| − | In MicroRumble
| |
| − | | |
| − | 84.75 APS
| |
| − | | |
| − | 97.95 PWIN
| |
| − | | |
| − | Improvements to random movement flattener
| |
| − | | |
| − | adapted flattener to take into account the robots velocity when deciding when to swap directions. Longer time before it can reverse at slow speeds, and a faster time to reverse at high speeds.
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=15TwArmvGV7bp6PUbBRfMkTHffCvmqjMK
| |
| − | | |
| − | ----
| |
| − | | |
| − | December 8, 2019, Basilite 0.12: Ranked 4 in MicroRumble, 12 in MiniRumble, 92 in RoboRumble
| |
| − | | |
| − | 84.86 APS
| |
| − | | |
| − | 98.15 PWIN
| |
| − | | |
| − | Reverted change in 0.10
| |
| − | | |
| − | https://drive.google.com/uc?export=download&id=1fYMpFL5XY2i1l9RQerIHyRn2d2_0nRwL
| |
For several times I see a learning gun bumping from one direction to another direction,
and yet each time an aim isn't satisfied, so no bullet is fired at all for a long time.
Actually this scenario disables the gun, until it gradually learns the new movement of the opponent.
However this behavior doesn't make sense. A prediction should be consistent,
at least for a short period of time (and until new information that denies previous assumptions appears),
or you're basically saying your own prediction is wrong, by changing mind frequently.
So, analog to human brains, could we just introduce an inertia to gun prediction process,
making it stick to previous predictions a little bit more, and as a side effect reducing noise?
E.g., introducing some rolling average, accumulating previous predictions with the newest ones.
This could be done at firing angle, but can also be done better, by doing the "accumulation"
(merging previous cluster with newest cluster, and weight them properly, or adding things up with VCS bins)
before the kernel density estimation.
