Scaling

Command:

Math -> Scaling of Data...

The command Math/Scaling of Data... (toolbar button ), provides some commonly used methods of data scaling. After clicking this command the following window appears:

The scaling can be applied either to the entire data set, to all columns or all rows, or to marked data only (again column- or row-wise). The type of marked data (type A or B, red or blue) is of no importance. The current state of the data table is copied into an auxiliary table when calling the scaling command, thus enabling the user to undo any changes as long as the scaling window is open.

The following scaling modes are available:

mean=0.0	Mean centering of the data.
mean=0.0, stddev=1.0	Scale to zero mean and unit variance (standardize).
sum = A	Scale to a constant sum. The value of constant has to be specified as parameter A.
max(|x|) = A	Divide all values by the maximum of the absolute values and multiply them by the parameter A.
min = A / max = B	Scale to a range so that the minimum equals parameter A and the maximum equals B.

You can store the transformation equations as a script on the harddisk by ticking off the check box "create transformation script". The name of the generated script has to specified in the box at the right of the checkbox. If you do not specify a path, the script will be stored in the working directory of DataLab. The storing of the transformation script is especially useful if several data sets of the same origin have to be processed (e.g. training and test sets). In order to apply the script you have to start the script editor, and load and excute it.

Please note that the set of equations can only be applied to unknown data if the data matrix has the same structure (i.e. the number of variables/rows and their ordering have to match).

Hint:

The scaling window can also be called directly from the numeric table editor.