| |
 |
 |
|
Last update: February 19, 2014
|
|
|
 |
 |
|
|
Class Merging & Discretization
For classification and logistic regression the learning algorithms of GeneXproTools
require a binary response variable, such as {0,
1} or {-1, 1}. This means that response variables with more than two different classes or
values must be converted to binary (broadly defined as having two different numerical values,
as GeneXproTools supports different
class representations). Thus, combining the
support for categorical variables
with the tools for merging and discretizing
the response, datasets with multiple classes and
datasets with numerical responses (both continuous and discrete) can be used for creating
classification and logistic regression models.
For datasets with multiple classes GeneXproTools allows you to set the singled out class to
C1 for example, and then create models for the binomial classification
task {class C1, not class C1}. Then single out another class and
create models for it too, and so on until you've created models for all sub-tasks. The merging and discretization
of the response variable takes place in the Class Merging & Discretization Window.
On the other hand, for datasets with a numerical response (loosely defined as having more
than two different values), such as the output of a logistic regression model with continuous
probability values between [0, 1], you can use the discretization function of GeneXproTools and
easily convert the continuous output into a binary outcome, such as {0, 1} or {-1, 1},
by choosing 0.5 as the discretization threshold.
See Also:
Related Tutorials:
Related Videos:
|
|
|
|
|
