The arboretum procedure

The NEURAL Procedure

PERTURB_Statement__Perturbs_weights._Perturbing_weights_can_sometimes_allow_you_to_escape_a_local_minimum.__Category'>PERTURB Statement

Perturbs weights. Perturbing weights can sometimes allow you to escape a local minimum.

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

affect only that statement.

PERTURB weight-list / <RANDF=number >

<OUTEST=SAS-data-set>

<RANDIST=name>

<RANDOM=integer>

<RANLOC=number>

<RANSCALE=number>;

Required Argument

weight-list

List of weights to freeze.

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 layer ID

Options

RANDF=number

Specifies the degrees of freedom parameter for random numbers. See the 

Randomization Options

and Default Parameters

 table for values.

OUTEST=SAS-data-set

Specifies the output data set containing all the weights.

Default:

none

RANDIST=name

---

Specifies the type of distribution for random numbers. See the 

Randomization Options and

Default Parameters

 table for values.

RANDOM=integer

Specifies the random number seed.

Default:

0

RANLOC=number

Specifies the location parameter for random numbers. See the 

Randomization Options and Default

Parameters

 table for values.

RANSCALE=number

Specifies the scale parameter for random numbers. See the 

Randomization Options and Default

Parameters

 table for values.

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

---

The NEURAL Procedure

PRELIM Statement

Performs preliminary training to reduce the risk of bad local optima.The final weights and biases

in a trained network depend on the initial values. The PRELIM statement repeatedely trains a

network for a smal number of iterations (default 20) using different initializations. The final

weights of the best trained network are then used to initialize a subsequent TRAIN statement.

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

affect only that statement.

PRELIM integer

<ACCELERATE=number>

<DECELERATE=number>

<INEST=SAS-data-set>

<LEARN=number>

<MAXLEARN=number>

<MAX | MAXMOMENTUM=number>

<MINLEARN=number >

<MOM | MOMENTUM=number>

<OUTEST=SAS-data-set>

<PREITER=integer>

<PRETECH=name>

<PRETIME=number>

<RANDBIAS|NORANDBIAS>

<RANDOUT|NORANDOUT>

<RANDOM=integer>;

Required Argument

integer

Specifies the number of preliminary optimizations.

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

INEST=SAS-data-set

Specifies the input data set that contains some or all weights. Any weights in the INEST= data set

that have missing values are assigned values according to the RANDOM=, RANDOUT, and

RANDBIAS options, as well as the options that pertain to random number distributions that you

specify in the Random statements.

LEARN=number

Specifies the learning rate for BPROP or the initial learning rate for QPROP and RPROP.

Range:

number > 0

Default:

0.1

MAXLEARN=number

Specifies the maximum learning rate for RPROP.

Range:

number > 0

Default:

Reciprocal of the square root of the machine epsilon

MAXMOM | MAXMOMENTUM=number

Specifies the maximum momentum for BPROP.

Range:

number > 0

Default:

1.75

MINLEARN=number

Specifies the minimum learning rate for RPROP.

Range:

number > 0

Default:

Square root of the machine epsilon

MOM | MOMENTUM=number

Specifies the momentum for BPROP.

Range:

0   number < 1

---

Default:

For BPROP: 0.9; for RPROP: 0.1

OUTEST=SAS-data-set

Specifies the output data set that contains all the weights.

PREITER=integer

Specifies the maximum number of iterations in each preliminary optimization.

Default:

10

PRETECH | TECHNIQUE=name

Specifies the optimization technique. See 

TRAIN Statement

.

Default:

Same as TECH= in the TRAIN statement.

PRETIME=number

Specifies the amount of time after which training stops.

RANDBIAS | NORANDBIAS

Specifies whether to randomize output biases.

Default:

NORANDBIAS, which sets bias to the inverse activation function of the

target mean.

[NORANDBIAS overrides whatever you specify in the RANDOM

statement.]

RANDOM=integer

Specifies the random number seed.

Default:

0

RANDOUT | NORANDOUT

Specifies whether to randomize the output connection weights.

Default:

NORANDOUT, which sets weights to 0.

[NORANDOUT overrides whatever you specify in the RANDOM

statement.]

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

---