Default:
NOMINAL for character variables.
INTERVAL for numeric variables.
MESTA=number
Specifies the scale constant for M estimation.
Default:
Default value is computed from MESTCON to give consistent scale
estimates for normal noise.
MESTCON=number
Specifies the tuning constant for M estimation.
Default:
Huber: 1.5;
Biweight: 9;
Wave: 2.1*
SIGMA=number
Specifies the fixed value of the scale parameter.
Default:
By default, SIGMA is not used; but with OBJECT=LIKE, the scale
parameter is estimated.
STD=method
Specifies the standardization method, where method is
NONE|NO
Variables are not altered.
STD
Variables are rescaled to have a mean of 0 and a standard deviation of 1.
RANGE|RAN
Variables are rescaled to have a minimum of 0 and a range of 1.
MIDRANGE|MID
Variables are rescaled to have a midrange of 0 and a half-range of 1 (that is, a minimum of
-1 and a maximum of 1.
Default:
NO
Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.
The NEURAL Procedure
THAW Statement
Thaws frozen weights.
Category Action Statement - affects the network or the data sets. Options set in an action statement
affect only that statement.
THAW weight-list;
Required Argument
weight-list
List of weights to thaw.
Weight-list consists of 0 or more repetitions of:
wname --> wname-2 where:
wname
is the unit name, the layer ID, BIAS, or ALTITUDE
wname-2
is the unit name or the layer ID
Default:
All weights are thawed.
Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.
The NEURAL Procedure
TRAIN Statement
Trains the network.
Category Action Statement - affects the network or the data sets. Options set in an action statement
affect only that statement.
TRAIN OUT=
SAS-data-set
OUTEST=
SAS-data-set
OUTFIT=
SAS-data-set
<
ACCEL|ACCELERATE=
number>
<
DECEL|DECELERATE=
number>
<
DUMMIES | NODUMMIES>
<
ESTITER=
i>
<
LEARN=
number>
<
MAX|MAXMOMENTUM=
number>
<
MAXITER=
integer>
<
MAXLEARN=
number>
<
MAXTIME =
number>
<
MINLEARN=
number>
<
MOM|MOMENTUM=
number>
<
TECHNIQUE=
name>;
Options
ACCEL | ACCELERATE=number
Specifies the rate of increase of learning for the RPROP optimization technique.
Range:
number > 1
Default:
1.2
DECEL | DECELERATE=number
Specifies the rate of decrease of learning for the RPROP optimization technique.
Range:
0 < number < 1
Default:
0.5
DUMMIES | NODUMMIES
Specifies whether to write dummy variables to the OUT= data set.
Default:
NODUMMIES
ESTITER=i
i = 0
Writes only initial and final weights to the OUTEST=data set.
i > 0
Writes weights after every i iterations, as well as the final weights, to the OUTEST= data
set.
Default:
0
LEARN=number
Specifies the learning rate for BPROP or the initial learning rate for QPROP and RPROP.
Range:
number > 0
Default:
0.1
MAXITER=integer
Maximum number of iterations.
Default:
100 for TRUREG and LEVMAR
200 for QUANEW and DBLDOG
400 for CONGRA
1000 for BPROP, RPROP, and QPROP
MAXLEARN=number
Specifies the maximum learning rate for RPROP.
Range:
number > 0
Default:
1/MACSQRTEPS
MAXMOM | MAXMOMENTUM=number
Specifies the maximum momentum for BPROP.
Range:
number > 0
Default:
1.75
MAXTIME=number
Specifies the amount of time after which training stops.
Default:
7 days, that is, 604800 seconds
MINLEARN=number
Specifies the minimum learning rate for RPROP.
Range:
number > 0
Default:
MACSQRTEPS
MOM | MOMENTUM=number
Specifies the momentum for BPROP.
Range:
0 number < 1
Default:
For BPROP: 0.9
For RPROP: 0.1
OUT=SAS-data-set
Specifies the output data set that contains the outputs.
OUTEST=SAS-data-set
Specifies the output data set that contains the network weights.
OUTFIT=SAS-data-set
Specifies the output data set that contains the fit statistics.
TECHNIQUE=name
Specifies the optimization technique where name is one of the following:
TRUREG
Requests Trust region.
LEVMAR
Requests Levenberg-Marquardt.
DBLDOG
Requests Double dogleg.
QUANEW
Requests quasi-Newton.
CONGRA
Requests Conjugate gradient.
BPROP
Requests standard backprop (backpropagation), that is, a variation on an algorithm called