Tiêu đề 08_04_Metropolis Sampling.pdf ✅

pacmann.io © 2022 – Pacmann AI 22 Metropolis Sampling
pacmann.io © 2022 – Pacmann AI 23 Gibbs Sampling : Review Previously in gibbs sampling we need to find conditional posterior for each parameter, in dividing what is the next sampled parameter, it depends on the value of other parameters. The drawbacks itself the sample is always taken. Question : Is there any better choice ? The approach is that we can consider each parameter as independent, and we can reject some samples
pacmann.io © 2022 – Pacmann AI 24 Metropolis Sampling If we do our previous case using Metropolis sampling it can be like this 1. Set Initial Value for μ0 and σ0 2. For n iteration we do : a. Sample μt from Proposal Distribution , you can choose any proposal distribution , for example normal distribution μt ~ Normal(μt-1 ,width) Collect the sample μt
pacmann.io © 2022 – Pacmann AI 25 Metropolis Sampling If we do our previous case using Metropolis sampling it can be like this 2. For n iteration we do : b. Sample σt from Proposal Distribution , you can choose any proposal distribution , for example normal distribution σt ~ Normal(σt-1 ,width) Collect the sample σt