I am trying to calculate the coefficients of a logistic regression algorithm (PROC IML - see below). It works for most datasets I use, but sometimes it encounters a floating point overflow error which kills the rest of the program. I wish to be able to continue subsequent steps and output missing values for the coefficients if this condition occurs. The question is, is there a way to determine if this condition will occur and to interrupt the process if it does, so it does not kill the program. Or is there a way to stop SAS from killing the program even if it does encounter this error.

DO WHILE(diff > 0.000001 & iteration <= 10);

v = exp(x*beta_old) / (1 + exp(x*beta_old)) / (1 + exp(x*beta_old));

mu = exp(x*beta_old)/(1+exp(x*beta_old));

IF ALL(v) = 1 THEN DO;

v2 = x*beta_old + (1/v)#(y - mu);

probmat = t(x#v#wt) * x;

rank = ROUND(TRACE(GINV(probmat)*probmat));

IF rank < NROW(beta) THEN DO;

diff = 0;

beta = REPEAT(.,NCOL(x),1);

END;

ELSE DO;

beta = inv(t(x#v#wt)*(x)) * t(x) * (v2#v#wt);

diff = SUM(abs(beta_old - beta));

beta_old = beta;

END;

END;

ELSE DO;

diff = 0;

beta = REPEAT(.,NCOL(x),1);

END;

iteration = iteration + 1;

END;

Thanks

> Pierre Felx > Methodologist > pierre.felx@statcan.ca <mailto:pierre.felx@statcan.ca> > 11I - R.H. Coats > Tunney's Pasture > Ottawa, Ontario > K1A 0T6 > > Phone: (613) 951 0075 > Fax: (613) 951 1462 >