Namularbre/tp2-iaavancee
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'v1.0.0'
All checks were successful
SonarQube Scan / SonarQube Trigger (push) Successful in 23s
Details

Browse Source 
# Conflicts:
#	ex2.py
This commit is contained in:
Namu
2025-10-03 13:59:51 +02:00
parent fecea4f5a0 311d9ee271
commit e92d445afc
1 changed files with 11 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
ex2.py
View File

@@ -5,7 +5,6 @@ import gymnasium as gym
import torch import torch
import torch.nn as nn import torch.nn as nn
import torch.optim as optim import torch.optim as optim
import matplotlib.pyplot as plt
class DQN(nn.Module): class DQN(nn.Module):
@@ -18,10 +17,11 @@ class DQN(nn.Module):
""" """
super().__init__() super().__init__()
self.net = nn.Sequential( self.net = nn.Sequential(
nn.Linear(n_states, 64), nn.ReLU(), nn.Linear(n_states, 128), nn.ReLU(),
nn.Linear(64, n_actions) nn.Linear(128, n_actions)
) )
def forward(self, x): def forward(self, x):
""" """
@@ -38,7 +38,7 @@ def epsilon_greedy(epsilon: float, s, policy_net: DQN, n_actions: int) -> int:
return torch.argmax(policy_net(s)).item() return torch.argmax(policy_net(s)).item()
def train_and_save(weights_path="cartpole_dqn.pth", episodes=2_000, update_target_every=20) -> DQN: def train_and_save(weights_path="cartpole_dqn.pth", episodes=2_000, update_target_every=20):
env = gym.make('CartPole-v1') env = gym.make('CartPole-v1')
n_states, n_actions = env.observation_space.shape[0], env.action_space.n n_states, n_actions = env.observation_space.shape[0], env.action_space.n
@@ -47,16 +47,15 @@ def train_and_save(weights_path="cartpole_dqn.pth", episodes=2_000, update_targe
target_net.load_state_dict(policy_net.state_dict()) # same weights at start target_net.load_state_dict(policy_net.state_dict()) # same weights at start
target_net.eval() target_net.eval()
optimizer = optim.Adam(policy_net.parameters(), lr=1e-2) #1e-3 optimizer = optim.Adam(policy_net.parameters(), lr=1e-3)
gamma = 0.99 # discount factor gamma = 0.99 # discount factor
epsilon = 1.0 # Fréquence d'exploration initiale epsilon = 1.0 # Fréquence d'exploration initiale
eps_min = 0.05 # Fréquence d'exploration minimale eps_min = 0.01 # Fréquence d'exploration minimale
eps_decay = 0.995 # Facteur de réduction d'epsilon eps_decay = 0.999 # Facteur de réduction d'epsilon
memory = deque(maxlen=5_000) memory = deque(maxlen=100_000)
batch_size = 64 batch_size = 64
for ep in range(episodes): for ep in range(episodes):
print(f'Episode: {ep}/{episodes}')
s, _ = env.reset() s, _ = env.reset()
s = torch.tensor(s, dtype=torch.float32) s = torch.tensor(s, dtype=torch.float32)
done, total_r = False, 0 done, total_r = False, 0
@@ -109,12 +108,14 @@ def show(weights_path='cartpole_dqn.pth') -> None:
s, _ = env.reset() s, _ = env.reset()
s = torch.tensor(s, dtype=torch.float32) s = torch.tensor(s, dtype=torch.float32)
done = False done = False
total_r = 0.0
while not done: while not done:
a = torch.argmax(qnet(s)).item() a = torch.argmax(qnet(s)).item()
s_, r, done, _, _ = env.step(a) s_, r, done, _, _ = env.step(a)
total_r += r
s = torch.tensor(s_, dtype=torch.float32) s = torch.tensor(s_, dtype=torch.float32)
env.close() env.close()
print('Demonstration finished.') print(f'Demonstration finished. {total_r:.1f}')
if __name__ == '__main__': if __name__ == '__main__':