Timeline for $\gamma^t$ in REINFORCE update (Sutton-Barto RL book Exercise 13.2)
Current License: CC BY-SA 4.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Sep 12, 2023 at 20:27 | comment | added | Procuste | So yes I agree with what you've done, but like you I'm not 100% sure. As pointed out in the paper linked and in towardsdatascience.com/…, I think it's something a lot of people overlook | |
| Sep 11, 2023 at 19:05 | comment | added | cfml | With $\mathbb{E}_\pi$ I meant the on-policy undiscounted expectation. Since $\mu_\gamma$ is the discounted distribution, I believe we are left with a factor of $\gamma^t$ when rewriting as $\mathbb{E}_\pi$. However, this part is what I was (and still is) the most unsure about. | |
| Sep 9, 2023 at 19:41 | history | edited | Procuste | CC BY-SA 4.0 |
added 482 characters in body
|
| Sep 9, 2023 at 15:57 | history | edited | Procuste | CC BY-SA 4.0 |
added 6 characters in body
|
| Sep 9, 2023 at 15:57 | history | edited | Procuste | CC BY-SA 4.0 |
added 160 characters in body
|
| Sep 9, 2023 at 15:56 | review | Late answers | |||
| Sep 16, 2023 at 15:43 | |||||
| Sep 9, 2023 at 15:39 | history | edited | Procuste | CC BY-SA 4.0 |
added 57 characters in body
|
| S Sep 9, 2023 at 15:38 | review | First answers | |||
| Sep 23, 2023 at 15:41 | |||||
| S Sep 9, 2023 at 15:38 | history | answered | Procuste | CC BY-SA 4.0 |