Comments (2)
I notice that, during training, the RSSM states are always initialized from the states of the previous iteration(code). As far as I understand, the training data are randomly sampled from the replay, and the initial states seem not to be transferrable from one iteration to the next. Then why does dreamerv3 initialize the state in this way?
As from my understanding of the paper and code, random trajectories are sampled from the replay buffer and each state of the trajectory is accessed with a queue in line 75 of your linked file.
But I am not sure if state shouldn't be set to state = [None]
after each call of train_step
like it is done at the beginning of training. Otherwise the state is set to the last outcome of the previous trajectory batch for each subsequent call of train_step
and I am not sure if this is the correct behaviour. Or was that your inital question?
Moreover, I realize that Dreamerv3 learns the initial states, but I'm quite confused about how they get updated. I'm wondering whether the state initialization mentioned above somehow benefits the learnable states? If so, how?
Regarding your second question: The initial state is state = [None]
and is then internally updated by the RSSM. The learning is based on equation 4 of the paper.
from dreamerv3.
Hi, the replay buffer sets is_first = True at the first time step of each replayed sequence so that the RSSM resets itself at the beginning of each batch. The implementation is like that because it would also allow different replay schemes where batches are consecutive for training with truncated backprop through time.
from dreamerv3.
Related Issues (20)
- Ask about jit in ninjax.py HOT 1
- (Maybe) a small bug in replay buffer implementation HOT 1
- Lambda Return Calculation Bug? HOT 2
- Checkpoint loading - memory doubles HOT 1
- [Question] How is symlog used in training?
- Dockerfile CUDA Image Version, COPY Path, and Execution Instructions Need Updates HOT 5
- Calling JAXAgent train gets stuck if using larger image sizes (inside Ninjax) HOT 3
- Scaling encoder and decoder MLP sizes? HOT 3
- Meaning of `is_last` and `is_terminal` and their effect on the value function HOT 2
- [Question] Question about the implementation of the Critic in "Mastering Diverse Domains through World Models" HOT 2
- Addition of "tree_map(np.asarray, value)" to "_convert_inps" function in "JAXAgent" HOT 1
- Dict based action space HOT 3
- Extremely large gradient and vanishing images when using jax.precision=float32 HOT 2
- Weight Initialization HOT 1
- Scaling laws in low-data regime
- How to optimize trade-offs in Scaling Laws HOT 1
- DreamerV3-Obstacle Tower HOT 1
- Crafter-model size HOT 1
- details about SAC used in the paper 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 dreamerv3.