Dirichlet distributions in JMP (for Bayes prior)

stephen_pearson

Hi,

 

I am trying some Bayesian bootstrapping and would like to do it in JMP Pro. One of the prior distibutions I would like to use is Dirichlet. I was thinking of using the column data type Expression to hold vectors of the weights and using the Compress to Label Addin to index the unique values in the column so I know which row/column of the vector contains the correct weight to apply.

 

I can generate Dirichlet vectors using the R connector (below) but would like to do it in a formula column so I can take advantage of the Bootstrap feature in Pro. Any suggestions?

 

Names Default To Here( 1 );
R Init();

//input list, outputlist, Rcode
rc = R Execute(
{nSamples,pollResult},
{x},
	"
	library (DirichletReg)
	
	x <- rdirichlet(nSamples, c(728,584,138))"
);
Wait( 3 );
//get dataframe from R
resultMatrix = R Get(x);

//close connection
R Term();

//export results
dt = As Table( resultMatrix );

Milo

Hi Stephen,

 

You can paste the code below into a column formula, and it should sample from a Dirichlet. I'd recommend that you convince yourself that it's accurate before using it, since I didn't test it. I wrote it based off of https://en.wikipedia.org/wiki/Dirichlet_distribution#Random_number_generation

 

alpha = [1 2 3 4 5];
gams = Random Gamma(alpha)`;
gams/sum(gams); // values sampled from dirichlet with concentration parameters alpha 

 

Milo