This is a great video for people who don't really want to bother with traditional Q-learning because it's almost always impractical due to the size of the Q table, but they want to see it actually solve a problem with a simple, step-by-step writing and explanation of the algorithm. That's why I think this video is one of the best (applied) machine learning videos out there. Straight and to the point, no fluff like all the popular videos are. Simply writing the algorithm out and explaining what each part does. You don't explain how and why Q learning works, but that's obviously not the point of the video. The point of the video is to connect the theory to an application of it. Thanks for doing that so well!
What I learned in this video: 1. Discretizing continuous state space 2. Cleanly organizing code: training and rendering flag. 3. Displaying the learning result
@@revimfadli4666 I don't think there are major improvements between Gymnasium vs Gym. The support and bug fixes were transferred to another team, so they renamed Gym to Gymnasium.
I have been using a different coding structure using DQN to solve the mountain car v0 problem with discrete action. And the main problem was that every time I run this same code, I get different results, that is sometimes good and sometimes bad. What do you advice me. this problem is really embarrassing me. Thank you for this video and keep up the good work brother. I wish I can show you my code to discuss about it. But I think youtube does not allow us to do that. Any way I can't wait to see your response.
I have a video using DQN on MountainCar-v0 using discrete actions: ruclips.net/video/oceguqZxjn4/видео.html If you wrote your own DQN code, you can compare it to mind and check for mistakes: ruclips.net/p/PL58zEckBH8fCMIVzQCRSZVPUp3ZAVagWi You can use Stable Baselines3's DQN as well: ruclips.net/video/OqvXHi_QtT0/видео.html
Hi, thank you for producing this fantastic video! I'm curious about the version of the python and the libraries you used. I'm currently facing issues downloading the pickle library due to environment solving failures :(
Yes, the agent can fall into the hole when slippery is on even if it knows the best path. Think of "best path" as the path with the highest chance of success.
Hi, in case you're still looking for a Deep Q-Learning video, I've recently released a detailed one, but is on the FrozenLake environment: ruclips.net/video/EUrWGTCGzlA/видео.html
Yup, it’s the latest version of the Gymnasium library (as of recording, a few weeks ago). I have a link in the description for installation if you run into trouble. Note that the original Gym library is no longer maintained, the support has been moved to Gymnasium.
@@johnnycode Tell me how you speed up the training of the program? I'm trying on CarRacing-v2 and can't speed up the car. All races are in real time. so I will never train the model)
@@cr4zygleb621 You must turn off the animation during training by setting render mode to None. For example: env = gym.make("CarRacing-v2", render_mode=None)
Ready for Deep Q-Learning? ruclips.net/video/EUrWGTCGzlA/видео.html
This is a great video for people who don't really want to bother with traditional Q-learning because it's almost always impractical due to the size of the Q table, but they want to see it actually solve a problem with a simple, step-by-step writing and explanation of the algorithm. That's why I think this video is one of the best (applied) machine learning videos out there. Straight and to the point, no fluff like all the popular videos are. Simply writing the algorithm out and explaining what each part does. You don't explain how and why Q learning works, but that's obviously not the point of the video. The point of the video is to connect the theory to an application of it. Thanks for doing that so well!
What I learned in this video:
1. Discretizing continuous state space
2. Cleanly organizing code: training and rendering flag.
3. Displaying the learning result
thanks for gymnasium most youtube outdated tutorials is stuck with gym..
What are the major improvements compared to old gym?
@@revimfadli4666 I don't think there are major improvements between Gymnasium vs Gym. The support and bug fixes were transferred to another team, so they renamed Gym to Gymnasium.
@@johnnycode oh OK thanks, so outdated tutorials aren't that outdated after all?
@@revimfadli4666 That is correct😀
I have been using a different coding structure using DQN to solve the mountain car v0 problem with discrete action. And the main problem was that every time I run this same code, I get different results, that is sometimes good and sometimes bad. What do you advice me. this problem is really embarrassing me. Thank you for this video and keep up the good work brother. I wish I can show you my code to discuss about it. But I think youtube does not allow us to do that. Any way I can't wait to see your response.
I have a video using DQN on MountainCar-v0 using discrete actions:
ruclips.net/video/oceguqZxjn4/видео.html
If you wrote your own DQN code, you can compare it to mind and check for mistakes:
ruclips.net/p/PL58zEckBH8fCMIVzQCRSZVPUp3ZAVagWi
You can use Stable Baselines3's DQN as well:
ruclips.net/video/OqvXHi_QtT0/видео.html
❤
thanks dude this was very helpful.. subbed
Hi, thank you for producing this fantastic video! I'm curious about the version of the python and the libraries you used. I'm currently facing issues downloading the pickle library due to environment solving failures :(
Never mind, I solved it
Good concise video
is there a possibility that, due to the slippery flag, the agent chooses the best action (knowing the best path) but it falls in the hole?
Yes, the agent can fall into the hole when slippery is on even if it knows the best path. Think of "best path" as the path with the highest chance of success.
He just goes up n down pasting code, not teaching a thing, but the code seems to work. Need to pause a hundred times the ChatGPT-Voice of this guy
can you do the code using NN instead of Q Table in tensorflow or PyTorch please?
Hi, in case you're still looking for a Deep Q-Learning video, I've recently released a detailed one, but is on the FrozenLake environment: ruclips.net/video/EUrWGTCGzlA/видео.html
how you use pygame window here ?
What is the version of gym library of this video? Is it the latest version?
Yup, it’s the latest version of the Gymnasium library (as of recording, a few weeks ago). I have a link in the description for installation if you run into trouble. Note that the original Gym library is no longer maintained, the support has been moved to Gymnasium.
@theavgdev I have file not found error in line 12
When I change it to wb instead of rb permission error appears
how much time did it take to complete 5000 episodes?
I think it only took one minute.
@@johnnycode
Tell me how you speed up the training of the program? I'm trying on CarRacing-v2 and can't speed up the car. All races are in real time. so I will never train the model)
@@cr4zygleb621 You must turn off the animation during training by setting render mode to None. For example: env = gym.make("CarRacing-v2", render_mode=None)
where is the pkl ?
Run the code with training turned on, then you'll see a new pkl file.
your mail please
Please drop your questions here.