Dynamic Lift Reports
About Dynamic Lift Reports
The Dynamic Lift report
enables you to view a model's lift at a given point in time or
to compare the lift performance of several models on one chart. The
Dynamic Lift report creates the following charts:
-
Lift
-
Cumulative Lift
-
Percent Response
-
Cumulative Percent Response
-
Captured Response
-
Cumulative Captured Response
A Dynamic Lift report
can be created only for classification models with a binary target.
The charts that are
created for a Dynamic Lift report are also created in the Monitoring
Report, which creates multiple types of model comparison reports.
Before you can create a Dynamic Lift report, certain project and model
property settings must be set.
For PMML 4.0 and later
models, the Valid variable name option in
SAS Management Console must be set to Yes
by a SAS Model Manager administrator. For more information,
see SAS Model Manager: Administrator’s
Guide.
Verify Project and Model Property Settings
Verify Project Properties
Select the project name
and verify that the following project properties are set:
Training target variable
Specifies the name
of the target variable that was used to train the model. The model
must have the same training target variable as the project.
Target event value
Specifies the value
for the desired target variable event or state. For example, if a
model predicts when RESPONSE=YES, then the target event value is
YES.
Output event probability variable
Specifies the name
of the output event’s probability variable.
Verify Model Properties
For each model in the
Dynamic Lift report, open the model and verify the following properties
on the Model Properties page:
Properties
|
Property
|
Description
|
|---|---|
|
Target variable
|
Specifies the name of
the target variable. For example, if a model predicts when RESPONSE=YES,
then the target variable is
RESPONSE.
|
|
Score code
type
|
Specifies whether the
score code runs using a DATA step fragment or SAS code that is not
a DATA step fragment.
|
Note: Dynamic Lift reports are
not applicable to models whose Score code type property
has a value of PMML. For PMML 4.0 and later, a Dynamic Lift report
can be created for a PMML model whose Score code type is
DATA step.
Create a Dynamic Lift Report
After ensuring that
the appropriate project and model properties have been set, create
the report.
To create a Dynamic
Lift report:
-
Specify the Prior probability. The prior probability is the proportion of event observations to the total observations in the whole population. In this case, the whole population is the entire train table. Specify a value for the prior probability to be used as the true event proportion when assessment values are computed for the lift of a model.
Copyright © SAS Institute Inc. All rights reserved.