Comments (5)
I think you will need to define a policy that inherits from FeedFowardPolicy (or even ActorCriticPolicy), not Mlp or Cnn, because it will be a mix.
from rl-baselines-zoo.
Looking more closely at the gym miniworld envs, you don't need a custom policy in fact. You directly use a provided wrapper:
https://github.com/maximecb/gym-minigrid/blob/999599a412db112bc7efa9a0f72f8c315074f8bb/gym_minigrid/wrappers.py#L144
from rl-baselines-zoo.
In order to implement something similar to what is available as:
https://github.com/lcswillems/rl-starter-files
we need a custom policy anyway.
For examples with MiniGrid, given a Dict of image, mission text, and others -> we can either extract only image (ImgObsWrapper obs['image']), or flatten everything into a 1D vector (FlatObsWrapper, a MiniGrid-specific version of gym.wrappers.FlattenDictWrapper). Pretty much depending on gym and policies.
I locally made it able to run with ImgObsWrapper - CustomCnnPolicy:
class CustomGridCnnPolicy(BasePolicy): # as common.FeedForwardPolicy
"""
Assuming gym_minigrid.wrappers.ImgObsWrapper
"""
def __init__(self, *args, **kwargs):
print("CustomGridCnnPolicy(): {} {}".format(args, kwargs))
super(CustomGridCnnPolicy, self).__init__(*args, **kwargs,
cnn_extractor=rl_starter_cnn)
which doesn't converge faster than MlpPolicy. ;->
I am trying Cnn-Lstm according to one of the rl-starter-files models, but still ignoring mission text.
Then we can try FlatObsWrapper - another policy recovering image and text from a given vector.
from rl-baselines-zoo.
now in the roadmap of V3: DLR-RM/stable-baselines3#1
from rl-baselines-zoo.
closing this in favor of DLR-RM/stable-baselines3#243
from rl-baselines-zoo.
Related Issues (20)
- HER success_rate output [question]
- What is the best initial values for the parameters in the .yml? [question] HOT 4
- How is the trial value calculated for RL jobs? HOT 1
- update the baseline for new envs? [feature request] HOT 1
- Multithreaded training with SubprocVecEnv() not working HOT 11
- [feature request] Install Zoo as a package HOT 6
- Irreproducible zoo trials HOT 5
- BadZipFile when running PPO2. HOT 5
- Can we use it with Tenserflow 2.3.1 ? HOT 5
- Why during the evaluation VecNormalize stats arenot loaded from the current training stats? HOT 1
- [Question] Why do envs look different? HOT 3
- [Question] Why is ortho_init=False during hyperparameter searches? HOT 2
- [feature request] Hyperparameter optimization for pretrained agent
- Hyperparameter tuning using Custom env with RL-zoo and SB2 HOT 3
- SB3 v1.1 Breaking changes HOT 1
- [question] Transfer hyperparameters from optuna HOT 2
- [feature request]Add Mujoco-Gym since mujoco is free now HOT 2
- Missing (or can't find) Hyperparameters HOT 2
- Reproducible zoo result HOT 3
- I am getting ImportError in atari environment
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 rl-baselines-zoo.