Dear experts,
In the TMVA guide, section 9.1.4, it is indicated that variable ranking is not supported for Keras, yet in the output logs of the NN training using Keras there is a table of variable ranking, and the numbers actually makes sense.
Do you know if there was a change since the guide was written and now there is variable ranking? and if so, how does the ranking being calculated? or should I ignore the table?
In case it helps, here is the part of the log I refer to:
*: Ranking input variables (method unspecific)...*
IdTransformation : Ranking result (top variable is best ranked)
-
: --------------------------------------*
-
: Rank : Variable : Separation*
-
: --------------------------------------*
-
: 1 : x1 : 4.814e-01*
-
: 2 : x2 : 4.429e-01*
-
: 3 : x3 : 3.906e-01*
-
: 4 : x4 : 2.927e-02*
-
: 5 : x5 : 6.651e-03*
-
: 6 : x6 : 5.564e-03*
-
: 7 : x7 : 4.337e-03*
-
: 8 : x8 : 3.517e-03*
-
: 9 : x9 : 1.539e-03*
-
: 10 : x10 : 1.331e-03*
-
: --------------------------------------*
Many thanks in advance