If you want to create a new population, it's pretty easy. This will load nfig from the Example world4-1 folder and replay the best individual from generation 2259.
This will load nfig from the Example world1-1 folder and replay the best individual from generation 12. python smb_ai.py -replay-file "Example world1-1" -replay-inds 1213,1214.If you want to run any them to see how they perform, do: When this option is present, certain information regarding generation, individuals who have won, new max distance, etc. This is helpul if you have disabled the display but wish to know how your population is doing. If you wish to know when populations are improving or when individuals have won, you can set a debug flag. When this option is present, nothing will be drawn to the screen.
The emulator supports faster updates and because of that an option has been created to run this through only command line. Because of this, when the display is open (whether it's hidden or not) you can only run at the refresh rate of your monitor. You are unfortunately limited by the refresh rate of your monitor for certain things in PyQt. Because of that, you may accidentally have many more parents in your population gene pool than intended. NOTE: this is not supported in -load-inds since when you load it will be treated as an initially population and those individuals will be treated as parents. If you want to replay starting at a certain individual through the end of the folder, you can do and this will begin at best_ind_gen12 and run through the entire folder. This accepts the same syntax as -load-inds with one additional syntax. Indicate the /path/to/population/folder that you want to replay from. You must specify both of the below arguments: This is helpful if you want to watch particular individuals replay their run. Note that loading individuals only supports loading the best performing one from generations. If you want to specify certain ones you can do 5,10,15,12,901 where they are seperated by a comma and no space. If you want a range, you can do, where it will load individuals best_ind_gen2, best_ind_gen3. This dictates which individuals from the folder to actually load. Indicate the /path/to/population/folder that you want to load from. Whatever the case, you can do so by specify both of the below arguments: You may want to run experiments by combining individuals from different populations. This could be due to your computer crashing and you wanting to load part of an old population. This is also necessary to specify when creating a new population. If you want to specifically set it, you can do so. Normally if you load/replay an individual, the config file is taken from whatever folder you are loading/replaying. This sets the config file for all AI in that population. If you want to see all command line options, the easiest way is to run python smb_ai.py -h. python -m retro.import "c:\Users\chris\Downloads\Super Mario Bros. Make sure you run this on the folder, i.e.
Run python -m retro.import "/path/to/unzipped/super mario bros.Head on over to the ROM for Super Mario Bros.This contains information on the following: It was also able to learn: flagpole glitch with an enemy, walljump, and a fast acceleration. If you want to see my blog explaining how all of this works in great detail, go here. If you want to see a YouTube video describing this at a high level, and showcasing what was learned, take a look here.
0 kommentar(er)
