cjgeyer
commented
5 years ago No and I can't deal with it right now as I am in hospital but will take a look when I get out
On Sun, Oct 28, 2018, 2:49 PM Robert Link notifications@github.com wrote:
I am seeing the following odd behavior. When I run the sampler, like so:
out1 <- metrop(logpost, pv0, 10, scale=mcmc_scale, debug=TRUE)
I get a couple of good iterations, but then the sampler starts repeating the same proposal over and over:
out1$proposal [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [1,] 3.073006 0.7735492 2.280427 0.2525763 2.127533 285.1648 1.125720 11.24583 [2,] 3.146012 0.5470985 2.260854 0.2051526 2.255066 285.3296 1.251440 12.49165 [3,] 3.219018 0.3206477 2.241281 0.1577288 2.382599 285.4944 1.377160 13.73748 [4,] 3.079750 0.5285934 2.222736 0.3251265 2.038565 284.6061 1.331585 12.74362 [5,] 3.079750 0.5285934 2.222736 0.3251265 2.038565 284.6061 1.331585 12.74362 [6,] 3.079750 0.5285934 2.222736 0.3251265 2.038565 284.6061 1.331585 12.74362 [7,] 3.079750 0.5285934 2.222736 0.3251265 2.038565 284.6061 1.331585 12.74362 [8,] 3.079750 0.5285934 2.222736 0.3251265 2.038565 284.6061 1.331585 12.74362 [9,] 3.079750 0.5285934 2.222736 0.3251265 2.038565 284.6061 1.331585 12.74362 [10,] 3.079750 0.5285934 2.222736 0.3251265 2.038565 284.6061 1.331585 12.74362
You can see that starting with iteration 4, the proposal is the same every time.
Restarting from the point where the last one left off gives me a one more valid iteration, but the sampler quickly falls back into the same repetitive behavior:
out2 <- metrop(out1, pv0, 100, scale=mcmc_scale, debug=TRUE) out2$proposal [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [1,] 3.079750 0.5285934 2.222736 0.3251265 2.038565 284.6061 1.331585 12.74362 [2,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [3,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [4,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [5,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [6,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [7,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [8,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [9,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [10,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [11,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [12,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [13,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 [14,] 2.919562 0.5275253 2.213430 0.3326856 2.271547 286.5868 1.500605 12.14620 (...)
The same proposal then gets repeated for all of the remaining iterations.
I tried it one more time, starting from the end of the second run, and this time it was a few iterations before the repetition began:
out3 <- metrop(out2, pv0, 100, scale=mcmc_scale, debug=TRUE) out3$proposal[1:15,] [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [1,] 2.919562 0.5275253 2.213430 0.33268563 2.271547 286.5868 1.500605 12.14620 [2,] 3.127507 0.5089804 2.380828 -0.01134893 2.182718 286.1311 1.301835 12.62126 [3,] 3.100912 0.6157917 2.299748 0.24053632 2.345545 285.2590 1.331331 12.22602 [4,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [5,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [6,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [7,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [8,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [9,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [10,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [11,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [12,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [13,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [14,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 [15,] 3.082407 0.5776736 2.419722 0.02403483 2.273197 286.0605 1.381725 12.35563 (...)
Presumably R's RNG is getting mangled somehow, but nothing in my code touches anything related to the RNG. Do you have any idea what might be causing this behavior?
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/cjgeyer/mcmc/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/AAcdwPaUUrh80QIJPMDQbUYv781uDhc8ks5upgpDgaJpZM4X-Jfn .
rplzzz
I am seeing the following odd behavior. When I run the sampler, like so:
I get a couple of good iterations, but then the sampler starts repeating the same proposal over and over:
You can see that starting with iteration 4, the proposal is the same every time.
Restarting from the point where the last one left off gives me a one more valid iteration, but the sampler quickly falls back into the same repetitive behavior:
The same proposal then gets repeated for all of the remaining iterations.
I tried it one more time, starting from the end of the second run, and this time it was a few iterations before the repetition began:
Presumably R's RNG is getting mangled somehow, but nothing in my code touches anything related to the RNG. Do you have any idea what might be causing this behavior?