| 1 |
## Two approaches to set up an order-two Markov model |
| 2 |
|
| 3 |
## Works |
| 4 |
|
| 5 |
x[1] ~ dnNormal(0,1) |
| 6 |
|
| 7 |
delta ~ dnNormal(0,sqrt(1/3)) |
| 8 |
|
| 9 |
x[2] := x[1] + delta |
| 10 |
|
| 11 |
for (i in 1:98) { |
| 12 |
|
| 13 |
x[i+2] ~ dnNormal(2 * x[i+1] - x[i],1) |
| 14 |
|
| 15 |
} |
| 16 |
|
| 17 |
print("Take 1, calculating log-probability of x[1]") |
| 18 |
|
| 19 |
print(x[1].lnProbability()) |
| 20 |
|
| 21 |
clear() |
| 22 |
|
| 23 |
## Non-centered parameterization. Doesn't work (Rev will need to be force-closed) |
| 24 |
|
| 25 |
x[1] ~ dnNormal(0,1) |
| 26 |
|
| 27 |
delta ~ dnNormal(0,sqrt(1/3)) |
| 28 |
|
| 29 |
x[2] := x[1] + delta |
| 30 |
|
| 31 |
for (i in 1:98) { |
| 32 |
|
| 33 |
dx[i] ~ dnNormal(0,1) |
| 34 |
|
| 35 |
x[i+2] := dx[i] - x[i] + 2 * x[i+1] |
| 36 |
|
| 37 |
# # Use below print statement to watch how time to compute probability slows as model is built |
| 38 |
|
| 39 |
print(x[1].lnProbability()) |
| 40 |
} |
| 41 |
|
| 42 |
print("Take 2, calculating log-probability of x[1]") |
| 43 |
|
| 44 |
print(x[1].lnProbability()) |
| 45 |
|
| 46 |
moves[1] = mvSlide(x[1]) |
| 47 |
|
| 48 |
mymodel = model(x[1]) |
| 49 |
|
| 50 |
# mymodel.graph("temp.dot",verbose=true,bg="honeydew2") |
| 51 |
|
| 52 |
monitors[1] = mnScreen() |
| 53 |
|
| 54 |
mymcmc = mcmc(mymodel, moves, monitors) |
| 55 |
|
| 56 |
mymcmc.run(10) |
| 57 |
|
| 58 |
q() |