r - Why am I getting NA's for sigma in this gamlss call? -
the following question asked michael barton on cross validated , rejected because deemed computer question. regardless, think question interesting , wondering if can answered here.
the original post here.
i fitting gamlss model call:
gamlss(formula = image_name + random(biological_source_name) - 1, sigma.formula = biological_source_name - 1, family = "nbi", data = na.omit(data))
after 3 iterations error:
gamlss-rs iteration 1: global deviance = 3814 gamlss-rs iteration 2: global deviance = 7760 gamlss-rs iteration 3: global deviance = 7756 in digamma(y + (1/sigma)) : nans produced in digamma(1/sigma) : nans produced in digamma(y + (1/sigma)) : nans produced in digamma(1/sigma) : nans produced error in glim.fit(f = sigma.object, x = sigma.x, y = y, w = w, fv = sigma, : na's in working vector or weights parameter sigma
this suggests me estimated sigma of categorical predictors going 0. correct?
any suggestions on how go resolving this?
i contacted authors regarding this. issue negative binomial able model on dispersion, whereas data contains both under- , over-dispersed output variables, between different dependent variable groups. results in error sigma going 0.
the problem data underdispered. , sigma goes 0 , derivatives produced na’s. try fit double poisson dpo() in specific data set.
as recommended 1 of authors, distribution such double poisson allows fitting because standard deviation can modelled being both more or less mean. when using distribution, solved above problem me , able fit model.
gamlss(formula = metric ~ image_name + random(biological_source_name) - 1, sigma.formula = ~ biological_source_name - 1, family = "dpo", data = na.omit(data))
note use of dpo
in above example.
Comments
Post a Comment