topic Re: ERROR: Invalid Operation. ERROR: Termination due to Floating Point Exception in Statistical Procedures

ERROR: Invalid Operation. ERROR: Termination due to Floating Point Exception

mezerji — Mon, 13 Apr 2020 09:24:18 GMT

Hi,

I am running on a 64-bit version of SAS 9.4 from a remote system. I received these error messages:

ERROR: Invalid Operation.

ERROR: Termination due to Floating Point Exception

when trying to run the code below in 1000 simulation and after about 50 run.

proc nlmixed data=analysis_&sim ;
parms a0=0 a1=0 a2=0 b0=0 b1=0 b2=0 tau=0 sigma=1 sigma1=1 sigma2=1 nu=1;
bounds sigma>0, sigma1>0, sigma2>0, nu>0;
eta_zero_m=a0+a1*x1+a2*x2+z1;
p_zero_m=exp(eta_zero_m)/(1+exp(eta_zero_m)); /* probability y > 0 */
eta_cont_m=b0+x1*b1+b2*x2+z2;
mu_cont_m=exp(eta_cont_m);
delta_cont=eta_cont_m - (sigma**2/2);
mu_cont_p=exp(delta_cont);
if y=0 then ll=log(p_zero_m);
else ll=log(1-p_zero_m)+log(y)*(mu_cont_p*(1+nu)-1)-(mu_cont_p*(1+nu)+nu+2)*log(1+y) lgamma(mu_cont_p*(1+nu))
-lgamma(nu+2)+lgamma(mu_cont_p*(1+nu)+nu+2);
model y~general(ll);
random z1 z2 ~ normal([0,0],[sigma1**2,tau*sigma1*sigma2,sigma2**2])
subject=idnum;
ods output ParameterEstimates=estTT_&sim ConvergenceStatus=ConvergTT_&sim FitStatistics=FitTT_&sim;
run;
data matrix_TT;
set estTT_:;
output;
data conv_TT;
set ConvergTT_:;
output;
data fit_TT;
set FitTT_:;
output;
run;

Re: ERROR: Invalid Operation. ERROR: Termination due to Floating Point Exception

StatsMan — Mon, 13 Apr 2020 12:59:13 GMT

I would open a ticket with SAS Tech Support for this one. You can reach them through email (support@sas.com) or through the web (support.sas.com). Better to use those methods than the phone for a complicated problem like this ...

Re: ERROR: Invalid Operation. ERROR: Termination due to Floating Point Exception

Rick_SAS — Mon, 13 Apr 2020 15:21:37 GMT

You can use

%PUT &=SIM;

inside the macro loop so that you can find out which simulated data set is failing. Then you can use a WHERE clause to examine the data and try to run the PROC NLMIXED code just on that subset.

One possibility is that either eta_zero_m or eta_cont_m is too large (greater than 709) so that EXP() overflows.