15:39 For anyone watching in 2022 getting errors, you need python version 3.9.10 for mlagents-learn to work. If you have a later version, you can install the earlier version (without adding it to the path) and use virtualenv to make a virtual environment using the older python version. You may need to also do pip install torch once in the venv, as it is a dependency of mlagents. Took me a lot of googling to figure this out so hopefully this helps others :)
@LoneXeaglE3 жыл бұрын
Great video !! For the raycast radius what it does is help you detect small objects by making the radius bigger you will actually have more chances of getting a hit when going by a small object instead of just missing it.
@skizooooooooo4 жыл бұрын
Awesome tutorial! Thanks for sharing ^.^ I definitely need to try this
@DapperDinoCodingTutorials4 жыл бұрын
Have fun :)
@thealexguy14864 жыл бұрын
thanks for the awesome tutorial! How long did you leave the AIs running for before they learnt enough? I've left mine learning for 2 hours with 4.1mil steps and yet it still can barely make it past the first couple. All my other settings are basically the same as yours except for the rays I had 10 instead of 8. Am I missing anything, or does it really just take that long to learn?
@JasonPesadelo4 жыл бұрын
I made one flappy bird 2D ML agents - I put reward between the walls and penalties in the sky---it works pretty good, it takes 8 hours to trainning ---it is not perfect.
@lennyroche80894 жыл бұрын
Exactly what I needed! Thanks!
@squaresweets8574 жыл бұрын
Is there any way to train the bird inside the unity editor without having to use the python things. Like if I wanted to make a mobile game where you trained your own flappy bird in-game would I be able to do that? Great vid BTW :)
@ivanpolkanov32734 жыл бұрын
Hey bro. You probably have a tone ways but if math doesnt scare u mb u can check jaff heaton channel as variant for begginer.
@DapperDinoCodingTutorials4 жыл бұрын
I don't think Ml-Agents are designed to be trained outside of the editor. It "might" be possible but I'm not aware of any way to do it.
@DonGroutsVideos4 жыл бұрын
Great video! Is there any way to use this to determine if the level is too difficult or too easy? Then to figure out if the pipes need to be spawned further apart or closer together, or the gap to be increased or decreased?
@DapperDinoCodingTutorials4 жыл бұрын
I guess you can write some code in the OnEpisodeBegin method that takes the mean reward and then adjusts the pipe handler settings accordingly. I wouldn't recommend it though as it's up to you to make it as easy/difficult as you want and then assuming that you've set up the agent well and that the level is possible, with enough training and with a good enough simulation environment, it should become good enough to play "almost" perfectly.
@holstjt4 жыл бұрын
Thanks for the tutorial. Helped a lot! And as always, I learned that the details are important... I missed the part where the bird is sat with a collider as a trigger, so it kind of didn't work as a start.
@Cranky724 жыл бұрын
Great video and good explanation. thank you
@DapperDinoCodingTutorials4 жыл бұрын
Glad you enjoyed!
@samirehman45654 жыл бұрын
Amazing video man.
@DapperDinoCodingTutorials4 жыл бұрын
Glad you enjoyed it!
@JasonPesadelo4 жыл бұрын
Thanks for the tutorial, great help!!!!
@MienMerk4 жыл бұрын
Thankyou for sharing 😁
@DapperDinoCodingTutorials4 жыл бұрын
No problem 👍
@DailyFootballEditsDFE4 жыл бұрын
very nice
@DapperDinoCodingTutorials4 жыл бұрын
Very nice
@giangbuitam48684 жыл бұрын
i have a question. Can i use a ml-agents in real project ?
@gelis074 жыл бұрын
what?
@Sovreighn74 жыл бұрын
And now to train them for WAR! MWHAHAHAHA!
@DapperDinoCodingTutorials4 жыл бұрын
You've got the right idea :D
@dripyman6146 Жыл бұрын
Hey does anyone know how to get the trainer config