- Contemporary messages sorted: [ by date ] [ by thread ] [ by subject ] [ by author ] [ by messages with attachments ]

From: Alison Boeckmann <alisonboeckmann_at_fastmail.fm>

Date: Sat, 22 Nov 2014 09:56:07 -0800

Xinting and others are correct. The syntax of ACCEPT/IGNORE is

limited and cannot handle an expression such as $DATA ACCEPT=((A ==

1.OR.A == 2).AND.B<100) In the NM-TRAN syntax, OR is implied. NOT and

AND cannot be coded.

Instead, the ACCEPT must be negated as IGNORE to change AND to OR:

IGNORE=(.NOT.(A==1.OR.A==2)).OR.B>=100

But (.NOT.(A==1.OR.A==2)) is (A.NE.1.AND.A.NE.2), which cannot be

coded because of the .AND.

Katya's workaround takes advantage of the fact that A takes numeric

values and there are no values between 1 and 2.

Thus, (A.NE.1.AND.A.NE.2) can be implemented with an "or":

(A<1.OR.A>2)

Katya's code is

IGNORE=(B.GE.100) IGNORE=(A.GT.2) IGNORE=(A.LT.1)

The following is equivalent, because NM-TRAN combines all the

IGNORE conditions, whether in one IGNORE list or in multiple ones:

IGNORE=(B.GE.100,A.GT.2,A.LT.1)

The above discussion also applies if the initial ACCEPT had been

IGNORE. NM-TRAN parses the right sides identically.

The intended and workaround code would be

IGNORE=((A == 1.OR.A == 2).AND.B<100) ; intended

ACCEPT=(B.GE.100,A.GT.2,A.LT.1) ; workaround

What if A takes all integer values and the acceptable values of A

are 1 and 6? Then you need something like this, in which all the

conditions are OR'd together: IGNORE=(B.GE.100) IGNORE=(A.GT.6)

IGNORE=(A.LT.1)

IGNORE=(A==2,A==3,A==4,A==5) It could get messy, and I don't think it

can be done at all for a

continuous valued A. At some point, an "and" is needed and the

syntax doesn't allow it.

Instead, Bill Denny's idea would have to be used:

... add a column to the data set that makes the selection simpler

(e.g. set it to 1 if (A == 1 or A == 2) and B < 100). You might do this

outside NONMEM, e.g. Excel.

With NONMEM, it can be done in two separate runs, one to create a

new data set with new data item and a separate run to use IGNORE/ACCEPT

to omit records based on the value in the new data item.

I implemented a simple example. Prob1.ctl sets ACC to 0/1

and ACC is used for ACCEPT in prob2.ctl.

But the NM=TRAN abbreviated code syntax for logical expressions

is also limited, so ACC had to be set in several statements.

I think the code in prob1.ctl can be modified for *any* logical

condition, however complex.

The data file is a fragment.

Attached files:

testab.dat

prob1.ctl

prob2.ctl

I plan to put this in Guide IV NM-TRAN and Guide VIII Help for

NONMEM 74. Comments and suggestions are welcome, especially from

those of you who teach NONMEM classes. Can all this be explained

more clearly and succinctly?

There is a feature in NONMEM 74 that makes it possible

to combine both problems into one NONMEM run:

$TABLE record option "EXCLUDE_BY list".

If any variable in list is non-0, the record is excluded from the

table file. This is implemented as probe.ctl and runs with nm74a. The

nm74a alpha version is available on request (ask Bob Bauer),

but it is not tested, and is for exploratory use only.

It has other new features of interest.

======= testab.dat

C NO A B ID DV MDV

1 1 100 1 0 0

2 2 100 1 0 0

3 3 100 1 0 0

4 1 101 1 0 0

5 2 101 1 0 0

6 3 101 1 0 0

7 1 99 1 0 0

8 2 99 1 0 0

9 3 99 1 0 0

======= prob1.ctl

$PROB test of accept Run #1

; Example of implementation of

; $DATA ACCEPT=((A == 1.OR.A == 2).AND.B<100)

$INPUT NO A B ID DV MDV

$DATA testab.dat IGNORE _at_

$PRED

; Implements:

; ACC=((A == 1.OR.A == 2).AND.B < 100)

; This cannot be coded directly in NM-TRAN abbreviated code.

; Instead:

; acc=1 if (a==1.or.a==2).and..not.b>=100

ACC=0

IF (A.EQ.1.OR.A.EQ.2) ACC=1

IF (B.GE.100) ACC=0

; The model for this run is unimportant. Keep it simple.

Y=THETA(1)+ETA(1)+EPS(1)

$THETA 1

$OMEGA 1

$SIGMA .4

$TABLE NO A B ID DV MDV ACC

NOPRINT NOAPPEND NOHEADER FILE=prob1.tab

======= prob2.ctl

$PROB test of accept Run #2

$INPUT NO A B ID DV MDV ACC

; Must do a numeric test for ACC==1, because it appears in

; table file as 1.0000E+00

$DATA prob1.tab ACCEPT=(ACC.EQN.1)

$PRED

; the model in this problem should be the model

; used for simulation, analysis, etc.

Y=THETA(1)+ETA(1)+EPS(1)

$THETA 1

$OMEGA 1

$SIGMA .4

$TABLE NO A B ID DV MDV

NOPRINT NOAPPEND FILE=prob2.tab

===== probe.ctl

$PROB test of NONMEM 7.4 exclude_by feature

$INPUT NO A B ID DV MDV

$DATA testab.dat IGNORE _at_

$PRED ; sets ACC=0 if ((a==1.or.a==2).and.b<100)

ACC=0

IF (A.NE.1.AND.A.NE.2) ACC=1

IF (B.GE.100) ACC=1

; this is the model for both problems.

; It may be more complicated than this.

Y=THETA(1)+ETA(1)+EPS(1)

$THETA 1

$OMEGA 1

$SIGMA .4

; probe.tab has only those records that have ACC=0.

$TABLE NO A B ID DV MDV EXCLUDE_BY ACC

NOPRINT NOAPPEND NOHEADER FILE=probe.tab $PROB test of exclude_by #2

$INPUT NO A B ID DV MDV

$DATA probe.tab (10F12.0) NOOPEN

$THETA 1

$OMEGA 1

$SIGMA .4

; This is the problem that should have $ESTIMATION or

; other tasks using the selected portion of the data set.

--

Alison Boeckmann

alisonboeckmann_at_fastmail.fm

Received on Sat Nov 22 2014 - 12:56:07 EST

Date: Sat, 22 Nov 2014 09:56:07 -0800

Xinting and others are correct. The syntax of ACCEPT/IGNORE is

limited and cannot handle an expression such as $DATA ACCEPT=((A ==

1.OR.A == 2).AND.B<100) In the NM-TRAN syntax, OR is implied. NOT and

AND cannot be coded.

Instead, the ACCEPT must be negated as IGNORE to change AND to OR:

IGNORE=(.NOT.(A==1.OR.A==2)).OR.B>=100

But (.NOT.(A==1.OR.A==2)) is (A.NE.1.AND.A.NE.2), which cannot be

coded because of the .AND.

Katya's workaround takes advantage of the fact that A takes numeric

values and there are no values between 1 and 2.

Thus, (A.NE.1.AND.A.NE.2) can be implemented with an "or":

(A<1.OR.A>2)

Katya's code is

IGNORE=(B.GE.100) IGNORE=(A.GT.2) IGNORE=(A.LT.1)

The following is equivalent, because NM-TRAN combines all the

IGNORE conditions, whether in one IGNORE list or in multiple ones:

IGNORE=(B.GE.100,A.GT.2,A.LT.1)

The above discussion also applies if the initial ACCEPT had been

IGNORE. NM-TRAN parses the right sides identically.

The intended and workaround code would be

IGNORE=((A == 1.OR.A == 2).AND.B<100) ; intended

ACCEPT=(B.GE.100,A.GT.2,A.LT.1) ; workaround

What if A takes all integer values and the acceptable values of A

are 1 and 6? Then you need something like this, in which all the

conditions are OR'd together: IGNORE=(B.GE.100) IGNORE=(A.GT.6)

IGNORE=(A.LT.1)

IGNORE=(A==2,A==3,A==4,A==5) It could get messy, and I don't think it

can be done at all for a

continuous valued A. At some point, an "and" is needed and the

syntax doesn't allow it.

Instead, Bill Denny's idea would have to be used:

... add a column to the data set that makes the selection simpler

(e.g. set it to 1 if (A == 1 or A == 2) and B < 100). You might do this

outside NONMEM, e.g. Excel.

With NONMEM, it can be done in two separate runs, one to create a

new data set with new data item and a separate run to use IGNORE/ACCEPT

to omit records based on the value in the new data item.

I implemented a simple example. Prob1.ctl sets ACC to 0/1

and ACC is used for ACCEPT in prob2.ctl.

But the NM=TRAN abbreviated code syntax for logical expressions

is also limited, so ACC had to be set in several statements.

I think the code in prob1.ctl can be modified for *any* logical

condition, however complex.

The data file is a fragment.

Attached files:

testab.dat

prob1.ctl

prob2.ctl

I plan to put this in Guide IV NM-TRAN and Guide VIII Help for

NONMEM 74. Comments and suggestions are welcome, especially from

those of you who teach NONMEM classes. Can all this be explained

more clearly and succinctly?

There is a feature in NONMEM 74 that makes it possible

to combine both problems into one NONMEM run:

$TABLE record option "EXCLUDE_BY list".

If any variable in list is non-0, the record is excluded from the

table file. This is implemented as probe.ctl and runs with nm74a. The

nm74a alpha version is available on request (ask Bob Bauer),

but it is not tested, and is for exploratory use only.

It has other new features of interest.

======= testab.dat

C NO A B ID DV MDV

1 1 100 1 0 0

2 2 100 1 0 0

3 3 100 1 0 0

4 1 101 1 0 0

5 2 101 1 0 0

6 3 101 1 0 0

7 1 99 1 0 0

8 2 99 1 0 0

9 3 99 1 0 0

======= prob1.ctl

$PROB test of accept Run #1

; Example of implementation of

; $DATA ACCEPT=((A == 1.OR.A == 2).AND.B<100)

$INPUT NO A B ID DV MDV

$DATA testab.dat IGNORE _at_

$PRED

; Implements:

; ACC=((A == 1.OR.A == 2).AND.B < 100)

; This cannot be coded directly in NM-TRAN abbreviated code.

; Instead:

; acc=1 if (a==1.or.a==2).and..not.b>=100

ACC=0

IF (A.EQ.1.OR.A.EQ.2) ACC=1

IF (B.GE.100) ACC=0

; The model for this run is unimportant. Keep it simple.

Y=THETA(1)+ETA(1)+EPS(1)

$THETA 1

$OMEGA 1

$SIGMA .4

$TABLE NO A B ID DV MDV ACC

NOPRINT NOAPPEND NOHEADER FILE=prob1.tab

======= prob2.ctl

$PROB test of accept Run #2

$INPUT NO A B ID DV MDV ACC

; Must do a numeric test for ACC==1, because it appears in

; table file as 1.0000E+00

$DATA prob1.tab ACCEPT=(ACC.EQN.1)

$PRED

; the model in this problem should be the model

; used for simulation, analysis, etc.

Y=THETA(1)+ETA(1)+EPS(1)

$THETA 1

$OMEGA 1

$SIGMA .4

$TABLE NO A B ID DV MDV

NOPRINT NOAPPEND FILE=prob2.tab

===== probe.ctl

$PROB test of NONMEM 7.4 exclude_by feature

$INPUT NO A B ID DV MDV

$DATA testab.dat IGNORE _at_

$PRED ; sets ACC=0 if ((a==1.or.a==2).and.b<100)

ACC=0

IF (A.NE.1.AND.A.NE.2) ACC=1

IF (B.GE.100) ACC=1

; this is the model for both problems.

; It may be more complicated than this.

Y=THETA(1)+ETA(1)+EPS(1)

$THETA 1

$OMEGA 1

$SIGMA .4

; probe.tab has only those records that have ACC=0.

$TABLE NO A B ID DV MDV EXCLUDE_BY ACC

NOPRINT NOAPPEND NOHEADER FILE=probe.tab $PROB test of exclude_by #2

$INPUT NO A B ID DV MDV

$DATA probe.tab (10F12.0) NOOPEN

$THETA 1

$OMEGA 1

$SIGMA .4

; This is the problem that should have $ESTIMATION or

; other tasks using the selected portion of the data set.

--

Alison Boeckmann

alisonboeckmann_at_fastmail.fm

Received on Sat Nov 22 2014 - 12:56:07 EST

*
This archive was generated by hypermail 2.3.0
: Fri Sep 27 2019 - 16:42:16 EDT
*