Purpose of PROC DMNEURL
§
27
larger values than 5. There is one memory allocation which takes
Ò
Ñ
long
integer values, where
Ò
is the value specified with the NPOINT= option and
Ñ
is the value specified by the MAXCOMP= option. The following table lists
values of
Ò
Ñ
½¼¼¼¼¼¼
for specific combinations of
´Ò
ѵ
. This is the actual
memory requirement in Megabytes assuming that a long integer takes 4 bytes
storage.
n
m=3
m=4
m=5
m=6
m=7
m=8
5
0
0
0
0
0
2
7
0
0
0
0
3
23*
9
0
0
0
2
19*
172
11
0
0
1
7*
78
857
13
0
0
2*
19
250
3263
15
0
0*
3
46
683
10252
17
0*
0
6
97
1641
27903
19
0
1
10
188
3575
67934
The trailing asterisk indicates the default number of points for a given number
of components. Therefore, values larger than 8 for
in
MAXCOMP=i
are re-
duced to this upper range. It seems to be better to increase the value
of the
MAXSTAGE= option when higher precision is requested.
MAXFUNC=i :
¼
specifies an upper bound for the number of function calls in each optimization.
The default is MAXFUNC=500. Normally the default number of function calls
will be sufficient to reach convergence. Larger values should be used if the it-
eration history indicates that the optimization process was close to a promising
solution but would have needed more than the specified number of function
calls. Smaller values should be specified when a faster but suboptimal solution
may be sufficient.
MAXITER=i :
¼
specifies an upper bound for the number of iterations in each optimization.
The default is MAXITER=200. Normally the default number of iterations will
be sufficient to reach convergence. Larger values should be used if the itera-
tion history indicates that the optimization process was close to a promising
solution but would have needed more than the specified number of iterations.
Smaller values should be specified when a faster but suboptimal solution may
be sufficient.
MAXROWS=i :
½
specifies an upper bound for the number of independent variables selected for
the model. More specific, this is an upper bound for the rows and columns
of the X’X matrix of the regression problem. The default is
Ñ
ÜÖ
ÓÛ
×
¿¼¼¼
.
Note, that the
Ì
matrix used for the stepwise regression takes
Ò
Ö
ÓÛ
×
´Ò
Ö
ÓÛ
×
·
½µ
¾
double precision values storage in RAM. For the default
maximum size of
Ò
Ö
ÓÛ
×
¿¼¼¼
you will need more than
¿¼¼¼
£
½
¼¼
£
bytes
RAM, which is slightly more than 36 megabytes.
MAXSTAGE=i :
½
specifies an upper bound for the number of stages of estimation.
If
28
§
PROC DMNEURL: Approximation to PROC NEURAL
MAXSTAGE is not specified, the default is MAXSTAGE=5. When a missing
value is specified, the multistage estimation process is terminated
¯
if the sum-of-squares residual in the component selection process changes
by less than 1%
¯
or when an upper range of 100 stages are processed.
That means, not specifying MAXSTAGE= or specifying a missing value are
treated differently. Large values for MAXSTAGE= may result in numerical
problems: the discretization error may be too large and the fit criterion does no
longer improve and can actually become worse. In such a case the stagewise
process is terminated with the last good stage.
MAXSTPT=i :
½
specifies the number of values of the objective function inspected for the start
of the optimization process. Larger values than the default value may improve
the result of the optimization especially when more than three components are
used. The default is MAXSTPT=250.
MAXVECT=i :
¾
specifies an upper bound for the number of eigenvectors made available for
selection. The default is MAXVECT=400. Smaller values should be used
only if there are memory problems for storing the eigenvectors when too many
variables are included in the analysis. The specified value for MAXVECT=
cannot be smaller than that for MINCOMP=.
If the specified value of
MAXVECT= is larger than the value for MAXROWS= it is reduced to the
value of MAXROWS=.
MEMSIZ=i :
½
For interval targets and in a multiple stage process some memory consuming
operations are being performed. For very large data sets the computations may
significantly depend on the size of the available RAM memory for those com-
putations. By default MEMSIZ=8 specifies the availability of 8 mb of RAM
for such operations. Since other operations need additional memory not more
than 25 percent of the total amount of memory should be specified here. If
you are running out of memory during the DMNEURL run, you may actually
specify a smaller amout than the default 8 mb.
MINCOMP=i :
¾
specifies a lower bound for the number of components selected for predicting
the target in each stage. The default is MINCOMP=2. The specified value for
MINCOMP= cannot be larger than that for MAXCOMP=. The MINCOMP=
specification may permit the selection of components which otherwise would
be rejected by the STOPR2= option. PROC DMNEURL may override the
specified value when the rank of the
¼
matrix is less than the specified
value.
NOMONITOR :
supresses the output of the status monitor indicating the progress made in the
computations.