Hi,
The below code will create a measure that displays the entropy of an attribute. The lower the entropy value the better the separation between two classes. The example used here is from this webpage: http://www.learnbymarketing.com/481/decision-tree-flavors-gini-info-gain/
The table was loaded and then uppivoted. The attributes themselves have become values to filter on. The attachment contains the Power BI file where you can review the Power M code. Hope that you find this usefull.
Entropy = MINX (
ADDCOLUMNS (
VALUES ( 'Table (2)'[Value] ),
"v",
VAR _selVal =
CALCULATE ( SELECTEDVALUE ( 'Table (2)'[Value] ) )
VAR _PercALT =
CALCULATE ( [PercA], FILTER ( 'Table (2)', [Value] <= _selVal ) )
VAR _PercBLT =
CALCULATE ( [PercB], FILTER ( 'Table (2)', [Value] <= _selVal ) )
VAR _PercAGT =
CALCULATE ( [PercA], FILTER ( 'Table (2)', [Value] > _selVal ) )
VAR _PercBGT =
CALCULATE ( [PercB], FILTER ( 'Table (2)', [Value] > _selVal ) )
VAR _CountRowsLT =
COUNTROWS ( FILTER ( 'Table (2)', [Value] <= _selVal ) )
VAR _CountRowsGT =
COUNTROWS ( FILTER ( 'Table (2)', [Value] > _selVal ) )
VAR _CountRows =
COUNTROWS ( 'Table (2)' )
VAR _EntLT =
-1
* (
( _PercALT * LOG ( _PercALT, 2 ) )
+ ( _PercBLT * LOG ( _PercBLT, 2 ) )
)
VAR _EntGT =
-1
* (
( _PercAGT * LOG ( _PercAGT, 2 ) )
+ ( _PercBGT * LOG ( _PercBGT, 2 ) )
)
RETURN
( _CountRowsLT / _CountRows ) * _EntLT + ( _CountRowsGT / _CountRows ) * _EntGT
),
[v]
)
ClassA = COUNTROWS ( FILTER ( 'Table (2)', [Class] = "A" ) )
ClassAll = COUNTROWS ( 'Table (2)' )
ClassB = COUNTROWS ( FILTER ( 'Table (2)', [Class] = "B" ) )
PercA = [ClassA] / [ClassAll]
PercB = [ClassB] / [ClassAll]
Kind regards, Steve.
eyJrIjoiNGNjNWY4ZmMtNDcyNi00N2IyLTk0OWUtNDRiODk0ZDc1NTAxIiwidCI6IjQxOTQ4NWNlLTE4ZDYtNDE2Yi04ODAyLTQ0YmQ2ZGNiNTU1OCIsImMiOjl9&pageName=ReportSection
