Comments (3)
szrlee
commented on May 22, 2024
3
If you are going to implement the discrete version of SAC,
I suggest also making a try to save the probability of behavior policy \pi_t(a|s) on the sample trajectory and augment log(\pi/\pi_t) to the reward instead of the sampled entropy \log(\pi) term.
We found that multi-step divergence augmentation is better than (multi-step) augmented entropy formulation in our algorithmic framework (not the same as the SAC algorithm framework).
from tianshou.
Trinkle23897
commented on May 22, 2024
Yes, I'll implement soon.
from tianshou.
yhisme
commented on May 22, 2024
ok thank you very much!:)
from tianshou.
Related Issues (20)
- UnboundLocalError: cannot access local variable 'obs_space_dtype' in atari_wrapper.py HOT 1
- Revisit "warm-up" phase in examples
- Should we use torch.compile? HOT 2
- Should we use the new schedule-free optimizer? HOT 1
- Add the non-in-place counterpart of `Batch.to_torch`
- Provide a devcontainer, base GH actions off it
- Some issues regarding configuration parameters HOT 5
- Glad you agree with me on this ^^. I'm not sure whether anywhere in the code the retrieval of the slice with empty values is used. For me it's fine to completely remove it, however, many tests will need to be adjusted, as now many of them rely on this somehow weird retrieval mechanism.
- Revisit `Launcher` for starting multiple experiments HOT 1
- Adjust locations of setting the policy in train/eval mode HOT 1
- Change log is chaotic and partly uninformative HOT 2
- how to run RL using multi-nodes in cluster HOT 1
- Potential confusion about where start timesteps are collected in HL interfaces HOT 4
- Use Altair inside a notebook to display benchmark results
- Does Tianshou truly supports MARL out of the box? HOT 1
- Extend benchmark with mujoco v4 envs
- How can I make action sampling within the range specified by my environment when using onpolicy_trainer? HOT 6
- Document effects of the relations between buffer size, num workers and episode length
- Poetry update the torch versioned from cuda (2.0.1+cu118) to cpu (2.1.1) defaultly on Windows HOT 5
- [question] Why does Tianshou use a replay buffer in on-policy RL algorithms? HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from tianshou.