gaspersic/Gymnasium
1
0
Fork 0
mirror of https://github.com/Farama-Foundation/Gymnasium.git synced 2025-07-31 22:04:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix reinforce tutorial (#1337)

Browse Source
This commit is contained in:
JacopoBandoni
2025-03-23 16:12:39 +01:00
committed by GitHub
parent ba0fa45af7
commit ddf5c6f7b0
1 changed files with 3 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
docs/tutorials/training_agents/reinforce_invpend_gym_v26.py
View File

@@ -202,14 +202,11 @@ class REINFORCE:
deltas = torch.tensor(gs)
log_probs = torch.stack(self.probs)
# Calculate the mean of log probabilities for all actions in the episode
log_prob_mean = log_probs.mean()
log_probs = torch.stack(self.probs).squeeze()
# Update the loss with the mean log probability and deltas
# Now, we compute the correct total loss by taking the sum of the element-wise products.
loss = -torch.sum(log_prob_mean * deltas)
loss = -torch.sum(log_probs * deltas)
# Update the policy network
self.optimizer.zero_grad()
@@ -298,8 +295,7 @@ for seed in [1, 2, 3, 5, 8]: # Fibonacci seeds
# ~~~~~~~~~~~~~~~~~~~
#
rewards_to_plot = [[reward[0] for reward in rewards] for rewards in rewards_over_seeds]
df1 = pd.DataFrame(rewards_to_plot).melt()
df1 = pd.DataFrame(rewards_over_seeds).melt()
df1.rename(columns={"variable": "episodes", "value": "reward"}, inplace=True)
sns.set(style="darkgrid", context="talk", palette="rainbow")
sns.lineplot(x="episodes", y="reward", data=df1).set(