Difference between revisions of "LightR/Experiences learned from ScalarR"
(Initial commit) |
m (Robocode isn't a competition, it's a journey.) |
||
| (One intermediate revision by the same user not shown) | |||
| Line 1: | Line 1: | ||
-- From an Engineering Perspective | -- From an Engineering Perspective | ||
| − | === Don't design === | + | == Don't design == |
| + | |||
| + | === Don't over-design === | ||
Don't make your code suitable for every combination of strategies. Strategies are coupled by its nature, well-designed and highly decoupled code can get very messy because of one single awesome strategy. | Don't make your code suitable for every combination of strategies. Strategies are coupled by its nature, well-designed and highly decoupled code can get very messy because of one single awesome strategy. | ||
| Line 14: | Line 16: | ||
=== Accept messy code === | === Accept messy code === | ||
| − | If your math utility is messy, fix it. If your kd-tree is messy, fix it. But if | + | If your math utility is messy, fix it. If your kd-tree is messy, fix it. But if anything else in your robot is messy, accept it. Good strategies are born to be coupled, because this way it leverages information best. So just accept it. It must be messy. |
| + | |||
| + | == Do make tools == | ||
| + | |||
| + | === Make good virtualization tools === | ||
| + | Plot every bins. Plot every waves. Plot how you move tick by tick, and plot the predictions. Good strategy comes from good observations into data, so never mind plotting everything. | ||
| + | |||
| + | === Make good debugging tools === | ||
| + | I made [[RobocodeGL/2.0|my own robocode]]. Space key pause, long press slo-mo, this is what I do every day. I can't imagine how I can debug my robots without good tools. | ||
| + | |||
| + | === Do watch battles === | ||
| + | So with all the tools, you spend 80% of the time watching battles. When every single detail is made perfect, and every single algorithm is optimized to its best, you get the best robot. So just be patient. Robocode isn't a competition, it's a journey. | ||
| + | |||
| + | __NOTOC__ __NOEDITSECTION__ | ||
Latest revision as of 15:23, 29 January 2022
-- From an Engineering Perspective
Don't design
Don't over-design
Don't make your code suitable for every combination of strategies. Strategies are coupled by its nature, well-designed and highly decoupled code can get very messy because of one single awesome strategy.
Don't go modularized
Modularized design is your best friend. Modularized design is your archenemy. When it comes to a good strategy, exhaustive information sharing is the key to success, and yet the archenemy to modularized design. Don't go modularized.
Don't make abstractions
Abstraction is the core concept of software engineering. Good abstractions & standardized building blocks are essential to large scale software developing. But yet every single abstraction layer obfuscates the relationships between the top layer and the bottom layer, which makes optimization harder. A good strategy needs direct control into every single detail in the bottom layer, so don't make abstractions.
Do make really good abstractions
This does not conflict to the above. Abstractions are still essential because you don't make robots from sands. Kd-trees, math utilities, etc., these are really good abstractions and you don't make strategies into them. Decoupling is still good, so do make really good abstractions.
Accept messy code
If your math utility is messy, fix it. If your kd-tree is messy, fix it. But if anything else in your robot is messy, accept it. Good strategies are born to be coupled, because this way it leverages information best. So just accept it. It must be messy.
Do make tools
Make good virtualization tools
Plot every bins. Plot every waves. Plot how you move tick by tick, and plot the predictions. Good strategy comes from good observations into data, so never mind plotting everything.
Make good debugging tools
I made my own robocode. Space key pause, long press slo-mo, this is what I do every day. I can't imagine how I can debug my robots without good tools.
Do watch battles
So with all the tools, you spend 80% of the time watching battles. When every single detail is made perfect, and every single algorithm is optimized to its best, you get the best robot. So just be patient. Robocode isn't a competition, it's a journey.
