havBpNet-J Data Create

havBpNet:J

Intro    Network Classes Overview    Data Classes Overview    Packages

havBpNet:J Home Page

Creating a collection of patterns using the havTdata class

Each instance of havTdata will contain a data-set. A data-set contains some number of patterns where each pattern contains an input pattern and, optionally, an expected pattern. When creating an instance of havTdata, you must specify several things:

the number of patterns to be read in from the file,
the input pattern size (number of elements in each input pattern),
the expected pattern size (number of elements in each expected pattern),
the name of the file from which the patterns are to be read.

Consider the following code excerpt which creates two data sets - one for training data - a second for test data used to determine network performance during training.


//=========== pattern and layer sizes

    static int	number_of_test_patterns  = 104; // number of patterns in test data set
    static int	number_of_train_patterns = 104; // number of patterns in training data set
    static int	pattern_size             = 60;  // number of input elements in a pattern
    static int	expected_size            = 1;   // number of output (expected) elements in a pattern

//=========== data file names

    static String  infile   = new String("sonartrain.dat");  // input data file name
    static String  testfile = new String("sonartest.dat");   // input test data file name

             . . .

//========== create data sets for training and test data (creation reads in data)

    havTdata train_data = new havTdata(number_of_train_patterns, pattern_size, expected_size,
                                       infile, havSys.havFALSE, 1.0);
    havTdata test_data = new havTdata(number_of_test_patterns, pattern_size, expected_size, 
                                      testfile, havSys.havFALSE, 1.0);

By the end of this excerpt, both data sets have been created and filled from their respective files.

The havTdata class is smart enough to remember that havlog scaling has been applied so that later, as we ask for errors, we will be given errors in un-scaled form.

Creating a data-set for Input patterns only

The example above illustrates creation of havTdata data-sets for patterns containing both input pattern and expected-pattern components. To create a data-set for patterns without expected-pattern parts, the expected_size parameter should be set to zero.

Data Overview || Data Create || Data Scaling || Data Performance
Data Save & Restore || Data Presentation Order

Intro    Network Classes Overview    Data Classes Overview    Packages