01.11.2014 Views

Reference Manual HALCON/C

Reference Manual HALCON/C

Reference Manual HALCON/C

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

8 CHAPTER 1. CLASSIFICATION<br />

º SampKey (output control) .......................................................featureset long *<br />

Identification of the data set to train.<br />

Result<br />

read sampset returns H MSG TRUE. An exception handling is raised if it is not possible to open the file or it<br />

contains syntax errors or there is not enough memory.<br />

Parallelization Information<br />

read sampset is local and processed completely exclusively without parallelization.<br />

create class box<br />

Possible Predecessor Functions<br />

Possible Successor Functions<br />

test sampset box, T enquire class box, write class box, close class box,<br />

clear sampset<br />

See Also<br />

test sampset box, clear sampset, learn sampset box<br />

Tools<br />

Module<br />

set class box param ( long ClassifHandle, const char *Flag,<br />

double Value )<br />

Set system parameters for classification.<br />

set class box param modifies parameter which manipulate the training sequence while calling<br />

T learn class box. Only parameters of the classificator are modified, all other classificators remain unmodified.<br />

’min samples for split’ is the number of examples at least which have to train in one cuboid of this<br />

classificator, before the cuboid is allowed to divide itself. ’split error’ indicates the critical error. By its exceeding<br />

the cuboid divides itself, if there are more than ’min samples for split’ examples to train. ’prop constant’ manipulates<br />

the extension of the cuboids. It is proportional to the average distance of the training examples in this cuboid<br />

to the center of the cuboid. More detailed:<br />

extension ¢ prop = average distance of the expectation value.<br />

This relation is valid in every dimension. Hence inside a cuboid the dimensions of the feature space are supposed<br />

to be independent.<br />

The parameters are set with problem independent default values, which must not modified without any reason.<br />

Parameters are only important during a learning sequence. They do not influence on the behavior of<br />

T enquire class box.<br />

Default setting:<br />

’min samples for split’ = 80,<br />

’split error’ = 0.1,<br />

’prop constant’ = 0.25<br />

Parameter<br />

º ClassifHandle (input control) ...................................................class box long<br />

Classificator’s handle number.<br />

º Flag (input control) ............................................................string const char *<br />

Name of the wanted parameter.<br />

Default Value : ’split error’<br />

Value Suggestions : Flag ¾’min samples for split’, ’split error’, ’prop constant’<br />

º Value (input control) .......................................................number double / long<br />

Value of the parameter.<br />

Default Value : 0.1<br />

read sampset returns H MSG TRUE.<br />

Result<br />

Parallelization Information<br />

set class box param is local and processed completely exclusively without parallelization.<br />

<strong>HALCON</strong>/C <strong>Reference</strong> <strong>Manual</strong> / 2000-11-15

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!