havBpNet-J Data Presentation Ordering

havBpNet:J

Intro    Network Classes Overview    Data Classes Overview    Packages

havBpNet:J Home Page

Random Presentation Order

havTdata provides a mechanism for randomizing the presentation order of training patterns in a data set. Patterns in an instance of havTdata may be randomized (in terms of their order) by sending a Randomize() message to the data set. Subsequent calls to retrieve a pattern can then be made using the GetRandPat(...) message as illustrated in the following excerpt:


 static boolean show_selection_order = false; // controls train pass reporting
   .  .  .

 havTpat pat;    // working ref to current pat in use
   .  .  .

 //=============== training section of main training loop
     
 train_data.Randomize();           // Randomize the training pattern presentation order
 if (show_selection_order)
    train_data.ShowRandList();
     
 for (i=0; i<number_of_train_patterns; i++)  { // Loop for each training pattern
    pat = train_data.GetRandPat(i);
    in.PutValue(pat.GetPat());     // put input pattern in input layer
    hid.Cycle();                   // feed forward through the network
    out.Cycle();
    out.PutExpected(pat.GetExp()); // Put the expected response in the output layer
    out.Train();                   // Calculate error and update weights
    hid.Train();
 }   
 //============== end of training section of main training loop

You are responsible for re-randomizing the order of patterns in the data-set by sending the Randomize() message whenever the patterns’ order should be changed. In our applications, we typically do this after each pass through the training set when the training epoch (batch) size is smaller than the total number of patterns in the training set.

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

Intro    Network Classes Overview    Data Classes Overview    Packages