Suggestion: 3-argument cdf()
droodman opened this issue · comments
In my application, I want to accurately and efficiently compute
cdf(Normal(),x2) - cdf(Normal(),x1)
for general x1 and x2.
A complication: if x1 and x2 are both large, say >=10, then each cdf()
call returns 1 and the difference is 0. I get a more accurate result if I replace the above with
ccdf(Normal(),x1) - ccdf(Normal(),x2)
So in general, on each call, I need to check the values of x1 and x2 and decide which expression to use.
It seems to me that it would be good if the Distributions.jl interface were extended to this case. Then I could do
cdf(Normal(),x1,x2)
...and place my trust in a well-crafted implementation.
Oop, I see there is logdiffcdf()
.