Need help with checkpoint system

[Scurry8609]

Hey there,

So I was wondering if you have any documentation on getting the checkpoint system to work? I am using the V2 baseline_fast_v2.py file. I must be doing something wrong I'm sure.

Every time I try to continue where I left off the whole training process restarts .

Also is there a way to slow the emulation down a bit with said python file?

Thank you for the help.

[kardSIM]

run ./go_forever.sh it will always load the last checkpoint in runs folder

also you can modify

# put a checkpoint here you want to start from    
if sys.stdin.isatty():
    file_name = ""
else:
    file_name = sys.stdin.read().strip() #"runs/poke_26214400_steps"

in baseline_fast_v2.py

to slow down emulation i think you need to update action_freq in env_config

[PWhiddy]

Hey!

First part of what @kardSIM said is correct. action_freq will change how many game ticks between agent actions but not actually the emulator speed. To change emulation speed during training (not sure why you'd want to slow that part down tho) you can set pyboy emulation speed limit like here:

PokemonRedExperiments/v2/red_gym_env_v2.py

Line 121 in 44cfc4e

self.pyboy.set_emulation_speed(6)

or by adding a time.sleep() in the env step function

[Scurry8609]

Thank you all for the information, I'll try the steps out and report back, about the emulation speed, I just want to show my kids what's going on on the screen without The Flash (Ash) running a 1000 MPH lol

One question, can you provide an example on what I should be looking to add to this code. Is there something specific that should go there?

Sorry I just started working with all this yesterday and will probably have many questions. Thank you again for the help :)

put a checkpoint here you want to start from

if sys.stdin.isatty():1
file_name = ""
else:
file_name = sys.stdin.read().strip() #"runs/poke_26214400_steps"