Youtube-Code-Repository
ppo-implementation-details
Youtube-Code-Repository | ppo-implementation-details | |
---|---|---|
5 | 18 | |
844 | 558 | |
- | - | |
1.6 | 0.0 | |
10 months ago | about 2 months ago | |
Python | Python | |
- | GNU General Public License v3.0 or later |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
Youtube-Code-Repository
-
Overall loss in PPO, why does it matter?
In Phil tabor's implementation it calculates Actor and Critic loss separately (line 95+) and does not calculate equation 9.
-
Intrinsic Curiosity Module Pytorch multithreading cpu unable to fix seeds
I am working on an extension of this implementation https://github.com/philtabor/Youtube-Code-Repository/tree/master/ReinforcementLearning/ICM of the intrinsic curiosity module. It uses A3C(Actor -critic) as a policy and the ICM is a bolt on module.
-
PPO cannot play CartPole ?
A very good performance reference code, which convers in 200 episodes.
-
Rl algorithm implemented
Github code - https://github.com/philtabor/Youtube-Code-Repository/tree/master/ReinforcementLearning/PolicyGradient/DDPG/tensorflow2/pendulum
-
Lunar Lander using Deep Q-Learning
I was wondering why the code looked so familiar, not just the design, but even the syntax and names of functions. I went through these myself when I was learning: Youtube-Code-Repository/ReinforcementLearning/DeepQLearning at master ยท philtabor/Youtube-Code-Repository (github.com). Its by a YouTuber / Udemy course instructor that goes through the design and coding process from scratch. This is probably mostly lifted straight from that repo. He even has a video on doing the lunar lander example too.
ppo-implementation-details
-
low reward oscillations in PPO
Follow this for stable training in PPO: https://iclr-blog-track.github.io/2022/03/25/ppo-implementation-details/
-
PPO-clip: Computing gradient WITHOUT auto differentiation library, help please?
I am using this as implementation reference.
-
My PPO Algorithm is not learning, why?
I'm relying on this page/code, and getting some ideas from others like this, and trying to learn PyTorch along the way.
-
Overall loss in PPO, why does it matter?
I am using as base code the Phils Tabor Implementation and this site (and sometimes OpenAi repository), but I can't figure out how tensorflow/PyTorch knows which loss belongs to whom. When the loss is split, you create two separate tape.Gradient, but when overall loss is used, how can the model understand which part propagates and which doesn't?
-
What RL library supports custom LSTM and Transformer neural networks to use with algorithms such as PPO?
I am still working on it, but I used the ppo implementation of https://github.com/vwxyzjn/ppo-implementation-details and modifiy it. Fir transformer, i just implement with pytorch.
- My agent seems to be learning but not on a stable way
-
trying to reproduce baselines PPO2 atari breakout
yes I did read https://iclr-blog-track.github.io/2022/03/25/ppo-implementation-details/
- Noob question: why is this trivial problem not accordingly trivial to train? (PPO)
- Are there papers that do an empirical investigation on DRL hyperparameters?
- Understanding the effect of certain PPO hyperparameters on overall performance
What are some alternatives?
Respiratory-Disease-Coughing-Dataset-CNN - A collection of coughing audio files from Coswara, Coughvid, and Virufy as well as generated spectrograms for the use of machine learning
baselines - OpenAI Baselines: high-quality implementations of reinforcement learning algorithms
RL-Algorithms - This repository has RL algorithms implemented using python
recurrent-ppo-truncated-bptt - Baseline implementation of recurrent PPO using truncated BPTT
easytorch - EasyTorch is a research-oriented pytorch prototyping framework with a straightforward learning curve. It is highly robust and contains almost everything needed to perform any state-of-the-art experiments.
incubator - Collection of in-progress libraries for entity neural networks.
Quantsbin - Quantitative Finance tools
pyagents - Just our DRL playground.
minimalRL - Implementations of basic RL algorithms with minimal lines of codes! (pytorch based)
popgym - Partially Observable Process Gym