Hi everyone,
I used the Distributions package in order to generate a beta distribution, which works quite nicely:
using Distributions
d = Beta(10,15)
However - is it possible to integrate over an interval of the distribution? I’m looking for a Julia equivalent to R’s integrate function which allows something like:
integrate(function(p) dbeta(p, 10, 15), 0, 0.5)
Any hints on where to look are appreciated!
Regards,
Thomas
I don’t know any R so I’m not sure what algorithm the integrate function uses, but have you tried QuadGK.jl?
using QuadGK
integral, err = quadgk(x->pdf(d,x),0,0.5)
gives
julia> integral, err = quadgk(x->pdf(d,x),0,0.5)
(0.8462718725204469, 2.8282575934013288e-9)
Here, x->pdf(d,x) is an anonymous function that represents the probability density function (pdf in Distributions.jl).
This is what I get in Mathematica:
Thanks crstnbr,
that works like a charm!
Regards,
Thomas
This seems like a job for the cdf function:
julia> p = Beta(10, 15)
Beta{Float64}(α=10.0, β=15.0)
julia> cdf(p, 0.5) - cdf(p, 0)
0.8462718725204467
It’s way faster too:
julia> using BenchmarkTools
julia> @btime integral, err = quadgk(x->pdf($p,x),0,0.5);
8.000 μs (34 allocations: 912 bytes)
julia> @btime cdf($p, 0.5) - cdf($p, 0);
321.267 ns (0 allocations: 0 bytes)
Thanks, DNF,
the difference in speed in this case will definitely something I’ll have to consider.
Regards,
Thomas
To clarify, it’s not just faster, it should also be more accurate and more ‘correct’.
The cumulative distribution function is defined to be the integral of the probability density function. The QuadGK code is using numerical quadrature to approximate the CDF. It’s of course better to directly use the known analytical expression for the CDF, which is what the cdf function does.
As a bonus, you don’t need to import an extra package.
BTW, I’m not familiar with R, but I’m pretty sure this is how you should do it in R too.
Indeed, in R you would use pbeta(0.5, 10, 15), not a numerical integration.