Timeline for ABC Pseudo Marginal
Current License: CC BY-SA 4.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 16, 2021 at 13:25 | comment | added | Fiodor1234 | @Xi'an It is written in the book "Handbook of Approximate Bayesian Computation" in Chapter 9. | |
| Jan 16, 2021 at 13:06 | comment | added | Xi'an | Again, when using the unbiased approximation$$\hat p_\text{ABC}(\theta|y_\text{obs})\propto\sum_{i=1}^M \psi(y_i(\theta),y_\text{obs})$$one need design an effective manner to simulate from this approximation. Note that the $y_i(\theta)$ are the simulated pseudo-observations, which depend on the chosen value of $\theta$. This makes the method quite unpractical. | |
| Jan 16, 2021 at 12:08 | comment | added | Fiodor1234 | @Xi'an Is it the fact that indirectly in the second case you consider more that one data samples ? | |
| Jan 16, 2021 at 12:03 | comment | added | Fiodor1234 | @Xi'an The unbiased estimator property I assume that it is required in order to sample from the correct approximated posterior, but yes it's not the property that makes the second approach more desirable. But why someone should prefer the second one?? Mainly the only difference is the Monte Carlo estimation of the Likelihood. | |
| Jan 16, 2021 at 11:48 | comment | added | Xi'an | It is not because you have an unbiased estimator of $p_{ABC}(\theta|y_{obs})$ that you have an easy way to simulate from it. | |
| Jan 16, 2021 at 11:47 | history | edited | Xi'an | CC BY-SA 4.0 |
added 1 character in body
|
| Jan 15, 2021 at 16:56 | history | edited | Fiodor1234 | CC BY-SA 4.0 |
edited title
|
| Jan 15, 2021 at 13:31 | history | asked | Fiodor1234 | CC BY-SA 4.0 |