| Algorithm | Discrete Action Space | Continuous Action Space |
|---|---|---|
| A2C | ✔️ | ✔️ |
| REINFORCE | ✔️ | ✔️ |
| DQN | ✔️ | ❌ |
| ADQN | ✔️ | ❌ |
Below is an example tested using OpenAI's gym.
import torch.optim as optim
# Import the implemented module and the default neural network model
from grll.PG.models import ANN_V2
from grll.PG import A2C
# Environment
import gymnasium as gym
env = gym.make('CartPole-v0')
# Create the neural network
num_actions = env.action_space.n
num_states = env.observation_space.shape[0]
A2C_model = ANN_V2(num_states, num_actions)
# Create the optimizer
optimizer = optim.Adam(A2C_model.parameters(), lr=1e-4)
# Initialize the reinforcement learning class
advantage_AC = A2C(
env=env,
model=A2C_model,
optimizer=optimizer,
)
# Train the model
advantage_AC.train(trainTimesteps=1000000)
# Save the class
ADeepQLearning.save("./saved_models/test.obj")If you want to use a different algorithm, you can write it like this:
from GRLL.PG import REINFORCE
"""
Or
from GRLL.VB import DQN
from GRLL.VB import ADQN
"""If you have downloaded the pygame module, you can use the following two environments.
RacingEnv_v0: Receives the lengths of 5 sensors as the state and has four actions: right, left, acceleration, and brake.
MazeEnv_v0: Receives vector information of the entire map as the state and has four actions to move north, south, east, and west.