CS_ALearn/deep-rl-class
1
0
Fork 0
mirror of https://github.com/huggingface/deep-rl-class.git synced 2026-04-13 11:59:45 +08:00
Code Issues Projects Releases Wiki Activity

Add standardization of the returns to obtain more stable training of Reinforce

Browse Source
This commit is contained in:
chris1nexus
2022-10-17 04:08:13 +02:00
parent c9b9e4810a
commit 7c06423ca4
1 changed files with 13 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
unit5/unit5.ipynb
View File

@@ -673,8 +673,13 @@
" for t in range(n_steps)[::-1]:\n",
" disc_return_t = (returns[0] if len(returns)>0 else 0)\n",
" returns.appendleft( ) # complete here \n",
" \n",
"\n",
" \n",
" ## standardization of the returns is employed to make training more stable\n",
" eps = np.finfo(np.float32).eps.item()\n",
" ## eps is the smallest representable float, which is \n",
" # added to the standard deviation of the returns to avoid numerical instabilities\n",
" returns = torch.tensor(returns)\n",
" returns = (returns - returns.mean()) / (returns.std() + eps)\n",
" # Line 7:\n",
" policy_loss = []\n",
" for log_prob, disc_return in zip(saved_log_probs, returns):\n",
@@ -766,6 +771,12 @@
" disc_return_t = (returns[0] if len(returns)>0 else 0)\n",
" returns.appendleft( gamma*disc_return_t + rewards[t] ) \n",
" \n",
" ## standardization of the returns is employed to make training more stable\n",
" eps = np.finfo(np.float32).eps.item()\n",
" ## eps is the smallest representable float, which is \n",
" # added to the standard deviation of the returns to avoid numerical instabilities \n",
" returns = torch.tensor(returns)\n",
" returns = (returns - returns.mean()) / (returns.std() + eps)\n",
" # Line 7:\n",
" policy_loss = []\n",
" for log_prob, disc_return in zip(saved_log_probs, returns):\n",