Understand your Unity A.I. | Tensorboard

Подписаться 11 тыс.

Просмотров 6 тыс.

50% 1

In this video I'm going to show you how you can understand your Unity AI.
You will learn how the ML-Agents Tensorboard Charts look like and how to interpret them.
Please let me know if you encountered any problems. I'll try my best to help you out.
Discord: / discord
Patreon: / botacademy
Find me on:
Discord: / discord
Twitter: / bot_academy
Instagram: / therealbotacademy
Patreon: / botacademy
Download my Unity AI example:
github.com/Bot-Academy/BallJump
Credits:
13.42 - End
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Music: Ansia Orchestra - Hack The Planet
Link: • Ansia Orchestra - Hack...
Music provided by: MFY - No Copyright
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Contact: smarter.code.yt@gmail.com
Chapters:
00:00 Intro
00:33 What is Tensorboard & TensorFlow
01:07 Run Tensorboard
01:49 Tensorboard Structure
02:22 Graph: Cumulative Reward
02:50 Graph: Episode Length
03:35 Graph: Is Training
05:14 Graph: Policy Loss
05:54 Graph: Value Loss
07:31 Graph: Q1 & Q2 Loss
08:37 Graph: Beta / Entropy / Entropy Coeff
10:11 Graph: Epsilon
10:30 Graph: Extrinsic Reward
10:54 Graph: Extrinsic Value Estimate
12:22 Graph: Learning Rate
12:51 Where To Ask & What's Not Covered
13:18 How to support me
13:42 Outro

Опубликовано:

26 июл 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 20

@5252emre 3 года назад

i must say this ml agent is one of the best features in unity

@micabarshap 4 года назад

Really eager to hear form you - thanks

@grimoirworkshop6623 3 года назад

Thanks a lot for stuff you do, comes really handy

@livirineya7723 3 года назад

Great help, thx!

@rajeevjawaji4472 4 года назад

good job

@BlackSheeeper 4 года назад

nice

@orsimhon133 3 года назад

Thanks ! You said Entropy is how much the Agent is sure about his actions, so it is not suppose to increase over time if it like that ?

@BotAcademyYT 3 года назад

Exactly, ideally it should decrease. Here is a link to an answer about the entropy (for PPO): www.reddit.com/r/reinforcementlearning/comments/bse7l5/rl_ppo_alrorithm_understanding_value_loss_and/

@nurhamidaljaddhi2338 Год назад

Thank you for this aamazing video, really enjoy to follow. I already follow the instruction. However, I face a problem. When I start training the agent, the folder summaries cannot generated, so that I cannot move to visualize the result using tensorboard. Is there any steps I missing? Thank you.

@RyushoYosei Год назад

He says that it has changed and you need to point to the Result folder, instead of the summaries folder in newer versions

@nurhamidaljaddhi2338 Год назад

@@RyushoYosei Thank you for your response. It is work.

@alvaromachucabrena97 3 года назад

Hi, I have a question. In supervised learning the error is calculated between the difference of the network output and the actual output. Finally, that error propagates from right to left to update the network weights. My question is: Using PPO, what is the criteria for updating network weights? Because here there is no expected output, so I would like to know what is the error that is propagated for the update of weights.

@BotAcademyYT 3 года назад

Good question. I try to give you a short answer which leaves out a few information. In reinforcement learning, the network receives a state and action and what the network predicts is the expected value when taking this action in the state. The error is the real value of the state which is calculated by the immediate reward and future estimated rewards. The more often the agent visits a state, the more accurate the real state value gets because of more information about the future estimated reward. tldr; The error is the immediate reward * gamma * state value of the next state which itself changes during training There are a few great articles out there which explain it more detailed (just search for something like 'reinforcement learning error calculation'. I'll also make a few animated reinforcement learning videos for beginners this year.

@alvaromachucabrena97 3 года назад

@@BotAcademyYT Thanks a lot !

@alvaromachucabrena97 3 года назад

How can I get neural network's loss?