The NEURAL Procedure
DELETE_Statement__If_an_input_or_hidden_layer_is_not_contributing_the_predictive_ability_of_the_network,_you_can'>DELETE Statement
If an input or hidden layer is not contributing the predictive ability of the network, you can
remove that layer with a DELETE statement. The DELETE statement also removes all associated
weights.
Category Action Statement - affects the network or the data sets. Options set in an action statement
affect only that statement.
DELETE id-list;
Required Argument
id-list
Specifies the identifiers of layers to delete.
Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.
The NEURAL Procedure
FREEZE Statement
Normally during training, all weights are updated. If you freeze one or more weights, those
weights will retain their frozen value until a corresponding THAW statement is executed. Freezing
weights causes training to proceed faster and require less memory.
Category Action Statement - affects the network or the data sets. Options set in an action statement
affect only that statement.
FREEZE weight-list / < VALUE= number>
< EST= SAS-data-set>;
Required Argument
weight-list
List of weights to freeze.
Weight-list consists of 0 or more repetitions of:
wname --> wname-2 where:
wname
is a unit name, a layer identifier, BIAS, or ALTITUDE
wname-2
is a unit name or a layer identifier
Options
You can specify either VALUE= or EST= but not both. If neither option is specified, the weights are
frozen to their current values.
VALUE=number
Specifies the numeric value to which weights are to be frozen.
EST=SAS-data-set
Specifies the SAS data set containing the values to which weights are to be frozen.
Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.
The NEURAL Procedure
FREQ Statement
Specifies the frequency variable for training.
Category Variable Statement - specifies variables.
FREQ variable;
Options
variable
Specifies the frequency variable. (The frequency variable can contain integer and non-integer
values.)
Note: The FREQ variable is not required in the DATA= data set. The NEURAL procedure
searches for the name of the FREQ variable in the DATA=, VALIDATA=, and TESTDATA=
data sets. If the FREQ variable does not appear in any of these data sets, then the procedure issues
a warning but continues processing. For any data set that does not contain the FREQ variable, a
FREQ value of 1 is used for all observations.
Default:
The default value is taken from the DMDB metadata. The FREQ statement
overrides whatever is in the DMDB metadata. If a FREQ statement contains
no variable name, then no FREQ variable is used.
Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.
The NEURAL Procedure
HIDDEN Statement
You can specify as many HIDDEN statements as you want up to the limits imposed by computer
memory, time, and disk space. The hidden layers can be connected in any feedforward pattern
using CONNECT statements.
Category Action Statement - affects the network or the data sets. Options set in an action statement
affect only that statement.
HIDDEN integer / ID= name
< ACT= activation-function>
< BIAS|NOBIAS>
< COMBINE= combination-function>;
Required Arguments
integer
Specifies the number of units in the hidden layer.
ID=name
Specifies the identifier for the layer.
Options
ACT=activation-function
Specifies the activation function. See
ACTIVATION FUNCTIONS
.
Default:
For hidden units, the default activation function depends on the combination
function and on the number of units in the layer.
For COMBINE=ADD, the default is ACT=IDENTITY.
For COMBINE=LINEAR or EQSLOPES, the default is ACT=TANH.
For COMBINE=EHRADIAL, EQRADIAL, EVRADIAL,
EWRADIAL, or XRADIAL, the default is ACT=EXP if there is only
one hidden unit in the layer; otherwise the default is ACT=SOFTMAX.
BIAS | NOBIAS
Specifies whether to use bias.
Default:
BIAS
COMBINE=combination-function
Specifies the combination function. See
COMBINATION FUNCTIONS
.
Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.
The NEURAL Procedure
ACTIVATION FUNCTIONS
Activation Functions
FUNCTION
RANGE
FUNCTION
(OF NET INPUT t)
IDEntity
LINear
EXPonential
RECiprocal
SQUare
LOGistic
MLOgistic
SOFtmax
GAUss
SINe
COSine
ELLiott
TANh
Dostları ilə paylaş: |