---
library_name: stable-baselines3
tags:
- PandaReachDense-v2
- deep-reinforcement-learning
- reinforcement-learning
- stable-baselines3
model-index:
- name: A2C
  results:
  - task:
      type: reinforcement-learning
      name: reinforcement-learning
    dataset:
      name: PandaReachDense-v2
      type: PandaReachDense-v2
    metrics:
    - type: mean_reward
      value: -0.74 +/- 0.27
      name: mean_reward
      verified: false
---

# **A2C** Agent playing **PandaReachDense-v2**
This is a trained model of a **A2C** agent playing **PandaReachDense-v2**
using the [stable-baselines3 library](https://github.com/DLR-RM/stable-baselines3).

## Usage (with Stable-baselines3)
TODO: Add your code


```python
#install
!apt install python-opengl
!apt install ffmpeg
!apt install xvfb
!pip3 install pyvirtualdisplay
!pip install -r https://raw.githubusercontent.com/huggingface/deep-rl-class/main/notebooks/unit6/requirements-unit6.txt

# Virtual display
from pyvirtualdisplay import Display
virtual_display = Display(visible=0, size=(1400, 900))
virtual_display.start()

#imports
import pybullet_envs
import panda_gym
import gym

import os

from huggingface_sb3 import load_from_hub, package_to_hub

from stable_baselines3 import A2C
from stable_baselines3.common.evaluation import evaluate_policy
from stable_baselines3.common.vec_env import DummyVecEnv, VecNormalize
from stable_baselines3.common.env_util import make_vec_env

from huggingface_hub import notebook_login

#Define the environment called "PandaReachDense-v2"
env_id = "PandaReachDense-v2"

#Make a vectorized environment
env = make_vec_env(env_id, n_envs=4)

#Add a wrapper to normalize the observations and rewards. Check the documentation
env = VecNormalize(env, norm_obs=True, norm_reward=True, clip_obs=10)

#Create the A2C Model (don't forget verbose=1 to print the training logs).
model = A2C(policy = "MultiInputPolicy",
            env = env,
            gae_lambda = 0.9,
            gamma = 0.95,
            learning_rate = 0.001,
            max_grad_norm = 0.5,
            n_steps = 8,
            vf_coef = 0.4,
            ent_coef = 0.0,
            seed=11,
            policy_kwargs=dict(
            log_std_init=-2, ortho_init=False),
            normalize_advantage=False,
            use_rms_prop= True,
            use_sde= True,
            verbose=1)

#Train it for 1M Timesteps
model.learn(1_500_000)

#Save the model and VecNormalize statistics when saving the agent
model.save(f"a2c-{env_id}")
env.save(f"vec_normalize_{env_id}.pkl")

#Evaluate your agent
eval_env = DummyVecEnv([lambda: gym.make(env_id)])
eval_env = VecNormalize.load(f"vec_normalize_{env_id}.pkl", eval_env)

#  do not update them at test time
eval_env.training = False
# reward normalization is not needed at test time
eval_env.norm_reward = False

# Load the model
model = A2C.load(f"a2c-{env_id}")

#Evaluate model
mean_reward, std_reward = evaluate_policy(model, eval_env)
print(f"Mean reward = {mean_reward:.2f} +/- {std_reward:.2f}")
...
```