math - Numerical problems of computing exponential numbers (from Gaussian PDFs) -
i have list of small exponential values (for example exp(n)
, n<-300
), generated gaussian pdfs.
i want compute how each of them proportional sum, example following python-like pseudo-code does:
s = 0 # sum of values n in exponents: s += exp(n) n in exponents: k = exp(n)/s # want compute k each n
the problem is, since values of n
small, summation s
turns out 0 sometimes, , i'll division-by-zero error.
i know 1 thing can add constant value (say 300) n
prevent underflow, it'll cause overflow in other cases.
how can solve this?
i don't know whether i've expressed myself clearly, if of these doesn't make sense or grammar mistakes, please correct me. in advance.
as observed, can subtracting constant value m
n
.
to avoid overflow, not choose fixed m
, such m = - 300
. instead, choose m
maximum of n
. every normalized exponential value @ 1
, hence normalized sum @ number of exponents, should reasonably small.
Comments
Post a Comment