| 1 |
module Coal where |
| 2 |
|
| 3 |
import Probability |
| 4 |
import Data.Frame |
| 5 |
import Data.Array |
| 6 |
|
| 7 |
-- When we get polymorphism, maybe we can create an temporary array as a view on xs' |
| 8 |
-- I presume that using n directly is more efficient that using [length xs'] |
| 9 |
orderedSample n dist = do |
| 10 |
xs' <- sample $ iid n dist |
| 11 |
let xs = sort xs' |
| 12 |
return $ listArray n xs |
| 13 |
|
| 14 |
get_intervals (r:[]) (t:[]) tend = [(r,t,tend)] |
| 15 |
get_intervals (r:rs) (t1:ts@(t2:_)) tend = (r,t1,t2):get_intervals rs ts tend |
| 16 |
|
| 17 |
model (t1,t2) times = do |
| 18 |
let lam = 3; nMax = 30; a = 1; b = 365/200 |
| 19 |
|
| 20 |
-- n change points, n+1 intervals |
| 21 |
n <- min nMax <$> prior (poisson lam) |
| 22 |
|
| 23 |
-- even numbered order statistics over the interval [t1,t2] |
| 24 |
s' <- orderedSample (2*n+1) (uniform t1 t2) |
| 25 |
let s = t1:[s'!(2*i-1) | i <- [1..n]] |
| 26 |
|
| 27 |
-- n+1 rates |
| 28 |
g <- prior $ iid (n+1) (gamma a b) |
| 29 |
|
| 30 |
let intervals = get_intervals g s t2 |
| 31 |
|
| 32 |
observe times $ poisson_processes intervals |
| 33 |
|
| 34 |
return [ "n" %=% n, "s" %=% s, "g" %=% g, "intervals" %=% intervals] |
| 35 |
|
| 36 |
|
| 37 |
main dirname = do |
| 38 |
|
| 39 |
frame <- readTable "coal-times.csv" |
| 40 |
|
| 41 |
let times = frame $$ "time" :: [Double] |
| 42 |
|
| 43 |
return $ model (1851, 1963) times |