topic Re: proc nlin and outest statement in Statistical Procedures

proc nlin and outest statement

MaryA_Marion — Mon, 20 Mar 2017 02:32:15 GMT

SAS 9.4 TS LEVEL 1M3

Hello,I am having difficulty with outest statement and reference to new (?) dataset koko. Please see error message below. I think it should output the cov matrix of the parameter estimates? How to do? Could it be a semicolon or statement placement problem?

proc NLIN data = eight

BEST= 10

MAXITER= 100

METHOD= Gauss

CONVERGE=1.0E-6

LIST

ALPHA = 0.10;

parms Beta1 = 0.05 Beta2 = 4;

parms

Beta1 = 0.753362 to 1.0 by 0.01

Beta2 = 0.53261;

der.Beta1 = exp(beta2 * x);

/* OUTEST = KOKO(where = (_TYPE_ = "COVB")); */

model y = Beta1 * exp(Beta2 * x);

output

out = eightOut

p= y_hat

r= residual

stdr= SE_Resid

LCLM= LCL_Mean

UCLM= UCL_Mean ;

run; quit; title;

/* error msg

OUTEST = KOKO(where = (_TYPE_ = "COVB"))

ERROR: Cannot find a library containing subroutine KOKO.

ERROR: no dependent variable specified.

NOTE: The SAS System stopped processing this step because of errors.

WARNING: The data set WORK.EIGHTOUT may be incomplete. When this step was

stopped there were 0 observations and 0 variables.

WARNING: Data set WORK.EIGHTOUT was not replaced because this step was stopped.

OUTEST=SAS-data-set

specifies an output data set that contains the parameter estimates produced at each iteration.

See the section Output Data Sets for details.

Re: proc nlin and outest statement

PGStats — Mon, 20 Mar 2017 03:10:27 GMT

Option OUTEST belongs in the proc nlin statement.

proc NLIN data = eight
BEST= 10    
MAXITER= 100   
METHOD= Gauss 
CONVERGE=1.0E-6
LIST
ALPHA = 0.10
OUTEST = KOKO(where = (_TYPE_ = "COVB"));

Re: proc nlin and outest statement

MaryA_Marion — Mon, 20 Mar 2017 12:44:19 GMT

Hello,

thank you for your reply. There are two problems.
1. koko does not exist
2. can I run multiple parms statements that preferrably are labelled?

Sincerely,
Mary

2. title2 "Non-Linear Regression y ~ X";
proc NLIN data = eight
BEST = 10
MAXITER = 100
METHOD = Gauss
CONVERGE=1.0E-6
LIST
ALPHA = 0.10
OUTEST = KOKO(where = (_TYPE_ = "COVB"));
parms Beta1 = 0.05 Beta2 = 4;
parms
Beta1 = 0.753362 to 1.0 by 0.01
Beta2 = 0.53261;
parms
Beta1 = 0.5 to 2.0 by 0.1
Beta2 = 0.53261;
der.Beta1 = exp(beta2 * x);
model y = Beta1 * exp(Beta2 * x);
output
out = eightOut
p = y_hat
r = residual
stdr = SE_Resid
LCLM = LCL_Mean
UCLM = UCL_Mean ;
run; quit; title;

##- Please type your reply above this line. Simple formatting, no
attachments. -##

Re: proc nlin and outest statement

ballardw — Mon, 20 Mar 2017 19:43:40 GMT

@MaryA_Marion wrote:
Hello,

thank you for your reply. There are two problems.
1. koko does not exist

OUTEST creates an output set. If you use the name of an existing data set the procedure will overwrite the existing set.

Re: proc nlin and outest statement

PGStats — Tue, 21 Mar 2017 03:00:23 GMT

I don't think multiple parameter statements are allowed but you can list starting parameter values in a separate dataset. This should work:

title2 "Non-Linear Regression y ~ X";

data parameters;
parameter = "beta1"; estimate=0.05; output;
parameter = "beta2"; estimate=4; output;
parameter = "beta2"; estimate=0.53261; output;
parameter = "beta1";
    do estimate = 0.753362 to 1.0 by 0.01;
        output;
        end;
    do estimate = 0.5 to 2.0 by 0.1;
        output;
        end;
run;

proc NLIN data = eight
    BEST = 10
    MAXITER = 100
    METHOD = Gauss
    CONVERGE=1.0E-6
    LIST
    ALPHA = 0.10
    OUTEST = KOKO(where = (_TYPE_ = "COVB"));
parameters / pdata=parameters;
model y = Beta1 * exp(Beta2 * x);
output
    out = eightOut
    p = y_hat
    r = residual
    stdr = SE_Resid
    LCLM = LCL_Mean
    UCLM = UCL_Mean ;
run; quit; 
title;

(untested)

Re: proc nlin and outest statement

MaryA_Marion — Wed, 22 Mar 2017 15:10:49 GMT

All three did not appear on output. It may be safer to just run these one
at a time.

Mary

##- Please type your reply above this line. Simple formatting, no
attachments. -##

Re: proc nlin and outest statement

MaryA_Marion — Sat, 25 Mar 2017 12:47:11 GMT

Your approach of using a dataset for the parms was clever. However I did not feel comfortable with the output from SAS. I was not completely sure which set of parameters I was using. I am doing them one at a time now (3 runs of proc nlin).

As for koko, I have a SAS-win10 OS problem sometimes. After many tries I finally got koko to print out. proc nlin syntax is very sensitive.

Thank you for replying.

Mary