Leave Some Out

Leave Some Out
 
Leave Some Out (LSO) is used to validate a network.  Example rows or subsets of examples are removed from the training set and used for validating.  After the optional preparation that allows Controls to be set and some row types to be excluded all but the excluded example rows are changed to training types.  One subset or a random number of example rows is then selected and changed to validating types.  The training rows are then used to train the network without the currently selected subset of validating rows.  Training continues until a stop is hit.  It is important that the controls are set with a possible stop.  The simplest stop to use with LSO is a fixed number of cycles.  Validating stops are not needed as LSO runs its own validating cycle.  When the training stops the selected subset of validating rows is used for validation.  The process is repeated until all subsets have been used once.  After the process is completed a report is produced showing the validating and training results. The report also contains the Diagnostic Array data if it available.
 
1

Example Rows

1. Example RowsThe total number of example rows in the Grid after the optional preparation.
 
2

Sequential selection

2. Sequential selectionSelect subsets of example rows for validating starting at the first example row in each subset.  Every subset will contain the same number of rows.
 
3

Random selection

3. Random selectionSelect each subset of rows for validating starting at a randomly selected example row.  Every subset will contain the same number of rows.
 
 
4

Random row selection

4. Random row selectionSelect rows at random for validating up to the number set.
 
5

Selections

5. SelectionsThe number of times to select subsets or random rows for validating.
 
 
6

Maximum rows

6. Maximum rows
The maximum number of rows to select randomly for validating.
 
7

OK

7. OKClick to accept the settings and close the dialog.
 
 
8

Cancel

8. CancelClick to close the dialog.
 
 

Created with help of DrExplain