The arboretum procedure



Yüklə 3.07 Mb.

səhifə89/148
tarix30.04.2018
ölçüsü3.07 Mb.
1   ...   85   86   87   88   89   90   91   92   ...   148
: documentation
documentation -> From cyber-crime to insider trading, digital investigators are increasingly being asked to
documentation -> EnCase Forensic Transform Your Investigations
documentation -> File Sharing Documentation Prepared by Alan Halter Created: 1/7/2016 Modified: 1/7/2016
documentation -> Gaia Data Release 1 Documentation release 0

The NEURAL Procedure

CONNECT Statement

A network can be specified without any CONNECT statements. However, such a network will be

connected by default as follows. First all input layers are connected to the first hidden layer, then

each hidden layer except the last is connected to the next hidden layer. Finally, the last hidden

layer is connected to all output layers. If this particular architecture is not appropriate, use one or

more CONNECT statements to explicitly define the network connections.

Category Action Statement - affects the network or the data sets. Options set in an action statement

affect only that statement.



CONNECT id-list;

Required Arguments

id-list

Lists the identifiers of two or more layers to connect. The identifiers must have been previously

defined by the ID= option in an INPUT, a HIDDEN, or a TARGET statement. Each layer except

the last is connected to the next layer in the list. Connections must be feedforward. Loops are not

allowed.

For example, the following PROC NEURAL step connects the input layers to the output layer, the input

layers to the hidden units, and the hidden units to the output layer.

title 'Fully Connected Network';

proc neural data=mydata dmdbcat=mycat;

   input a b / level= nominal id=nom;

   input x z / level= interval id=int;

   hidden 2 / id= hu;

   target y / level=interval id=tar;

   connect int tar; 

   connect nom tar;

   connect int hu;

   connect nom hu;

   connect hu tar;

   train;

run;


Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.


The NEURAL Procedure

CUT Statement

If the weights corresponding to the connection between two layers are not contributing the

predictive ability of the network, you can remove that connection and the corresponding weights

by using a CUT statement.

Category Action Statement - affects the network or the data sets. Options set in an action statement

affect only that statement.



CUT id-list| ALL;

Options

You must specify either:



id-list

Specifies the identifiers of the layers to disconnect.



ALL

Disconnects all layers.

Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.



The NEURAL Procedure

DECISION Statement

Specifies information used for decision processing in the DECIDE, DMREG, NEURAL, and

SPLIT procedures. This documentation applies to all four procedures.

Tip: The DECISION statement is required for PROC DECIDE. It is optional for DMREG, NEURAL

and SPLIT procedures.



DECISION DECDATA= SAS-data-set <DECVARS=decision-variable(s)><option(s)>;

DECDATA=  SAS-data-set

Specifies the input data set that contains the decision matrix. The DECDATA= data set must

contain the target variable.

Note:   The DECDATA= data set may also contain decision variables specified by means of the

DECVARS= option, and prior probability variable(s) specified by means of the PRIORVAR=

option or the OLDPRIORVAR= option, or both.

The target variable is specified by means of the TARGET statement in the DECIDE, NEURAL,

and SPLIT procedures or by using the MODEL statement in the DMREG procedure. If the target

variable in the DATA= data set is categorical, then the target variable of the DECDATA= data set

should contain the category values, and the decision variables will contain the common

consequences of making those decisions for the corresponding target level. If the target variable is

interval, then each decision variable will contain the value of the consequence for that decision at

a point specified in the target variable. The unspecified regions of the decision function are

interpolated by a piecewise linear spline.  

Tip:

The DECDATA= data set may be of TYPE=LOSS, PROFIT, OR

REVENUE. If unspecified, TYPE=PROFIT is assumed by default. TYPE= is

a data set option that should be specified when the data set is created.



DECVARS=decision-variable(s)

Specifies the decision variables in the DECDATA= data set that contain the target-specific

consequences for each decision.

Default:

None


COST=cost-option(s)

Specifies numeric constants that give the cost of a decision, or variables in the DATA= data set

that contain the case-specific costs, or any combination of constants and variables. There must be

the same number of cost constants and variables as there are decision variables in the DECVARS=




option. In the COST= option, you may not use abbreviated variable lists such as D1-D3,

ABC--XYZ, or PQR:.



Default:

All costs are assumed to be 0.



CAUTION:

The COST= option may only be specified when the DECDATA= data set is of

TYPE=REVENUE.   

PRIORVAR=variable

Specifies the variable in the DECDATA= data set that contains the prior probabilities to use for

making decisions.

Tip:

In the DECIDE procedure, if PRIORVAR= is specified, OLDPRIORVAR=

must also be specified.

Default:

None


OLDPRIORVAR=variable

Specifies the variable in the DECDATA= data set that contains the prior probabilities that were

used when originally fitting the model.

Tip:

If OLDPRIORVAR= is specified, PRIORVAR= must also be specified.



CAUTION:

OLDPRIORVAR= is not allowed in PROC SPLIT.   

Default:

None


Copyright 2000 by SAS Institute Inc., Cary, NC, USA. All rights reserved.



Dostları ilə paylaş:
1   ...   85   86   87   88   89   90   91   92   ...   148


Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©genderi.org 2017
rəhbərliyinə müraciət

    Ana səhifə