TRAIN OUT=
SAS-data-set
OUTEST=
SAS-data-set
OUTFIT=
SAS-data-set
<
ACCEL|ACCELERATE=
number>
<
DECEL|DECELERATE=
number>
<
DECAY=
number>
<
DUMMIES|NODUMMIES>
<
ESTITER=
i>
<
LEARN=
number>
<
MAX|MAXMOMENTUM=
number>
<
MAXITER=
integer>
<
MAXLEARN=
number>
<
MAXTIME =
number>
<
MINLEARN=
number>
<
MOM|MOMENTUM=
number >
<
TECHNIQUE=
name>;
USE SAS-data-set;
Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.
The NEURAL Procedure
PROC NEURAL Statement
Invokes the NEURAL procedure.
PROC NEURAL <
option-list>;
Required Arguments
DATA=SAS-data-set
Specifies the DMDB-encoded input SAS data set containing the training data.
DMDBCAT=SAS-catalog
Specifies the DMDB catalog.
Options
GRAPH
Plots the objective function, validation error, and test error during training.
NETWORK=screen-specification
Constructs a network according to a description that was saved by using a SAVE statement during
a previous execution of the NEURAL procedure. screen-specification is the catalog entry that was
specified in the SAVE statement.
Default:
None
RANDOM=integer
Specifies the random number seed used in network weight initialization.
Default:
12345
CAUTION:
The weights and predicted outputs from the network cannot be reproduced when a 0
or negative RANDOM= value is specified. When a 0 or negative value is specified, the
system clock is used to generate the seed. The actual value of this seed will be
unavailable, and you lose control over the initialization of weights. Different
initializations will result in different final weights and predicted outputs for repeated
runs of the same set of NEURAL statements and same input data sets.
STOPFILE='file path name'
This option enables you to stop the NEURAL training when you are running a large job. Before
you invoke the NEURAL procedure, specify the file path name in the STOPFILE option. For
example, STOPFILE= "c:\mydir\haltneural". Initially, this file should not exist.The NEURAL
procedure checks for the existence of this file between iterations in the training process. When you
want to stop the job, create the specified file, and the NEURAL procedure will halt the training at
the current iteration. The file does not have to contain any contents.
TESTDATA=SAS-data-set
Specifies a data set used to compute the test average error during training. At selected iterations
(controlled by ESTITER=), each observation in the TESTDATA= data set is read in, scored using
the current network weight values, and the error computed. The average test error is then output to
the OUTEST= data set.
Note: This requires the TESTDATA= data set to contain the inputs and target variables.
VALIDATA=SAS-data-set
Specifies a data set used to compute the validation average error during training. At selected
iterations (controlled by ESTITER=), each observation in the VALIDATA= data set is read in,
scored using the current network weight values, and the error computed. The average validation
error is then output to the OUTEST= data set.
Note: This requires the VALIDATA= data set to contain the inputs and target variables.
Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.