Chapter 6.
Metropolis-Hastings algorithms
· Example 6.1 (pages 192-3) – Figure 6.1 (page 194) - Random walk Metropolis-Hastings algorithm to simulate from
π(θ) = fN(β;0,100)∏ fN(yi;50+170xi/(θ+xi),126), for θ in R.
Random walk proposal: q(θ,φ)= fN(φ;θ,0.01)
Dataset: y = velocity of an enzymatic reacton (in counts/min/min)
x = substrate concentration (in ppm)
x = (0.02,0.02,0.06,0.06,0.11,0.11,0.22,0.22,0.56,0.56,1.10,1.10)
y = (76,47,97,107,123,139,159,152,191,201,207,200)
Figure 6.1 shows the trajectory of the chain with initial value for θ equal to 0.4.
·
Example 6.2 (pages 199
and 202-3) – Figures 6.2 - 6.6 (pages 200-204) - Random
walk Metropolis-Hastings versus independence Metropolis-Hastings
The target density if a mixture of two 2-dimensional normal densities (Figure 6.2);
Part 1 of the example (page 199) implements the random walk Metropolis-Hasting algorithm:
Figure 6.3 – Chain paths for 6 combinations of initial values and tuning parameters;
Figure 6.4 – Chain autocorrelations for 6 combinations of initial values and tuning parameters.
Part 2 of the example (pages 202-204) implements the independence Metropolis-Hasting algorithm:
Figure 6.5 – Chain paths for 6 combinations of initial values and tuning parameters;
Figure 6.6 – Chain autocorrelations for 6 combinations of initial values and tuning parameters.
·
Example 6.4 - Random walk Metropolis-Hastings (single and block moves) and independence Metropolis-Hastings (block move)
Times to failure (f) of motorettes were tested at different temperatures (t).
There are 17 uncensored and 23 censored observations.
A constant prior is used for the regression parameters.
· Example 6.4 Part 1 :
(pages 210-1) – Figure 6.7 (page 212) -
Original regressor
· Example 6.4 Part 2:
(pages 216-7) – Figure 6.9 (page 218) -
Centered regressor
· Example 6.5 (pages
212-3) – Figure 6.8 (page 214) - Poisson model with change point
revisited. Metropolis step for
change point parameter.
· Exercise 6.9 (page 236) - Univariate version of Example 6.2 above.