Gradient Boosted Trees does not report attribute weights correctly
UnicornIn the attached example, the attribute weights port outputs:
| J.TIPO_DE_ESFERASTER | 6.545000076293945 |
| P.PERFORANTE | 5.506167411804199 |
| F.TIPO_DE_EXOSTILO | 3.6711277961730957 |
| C.TIPO_ESPICULA_PRINCIPAL_ESTRONGILOXA | 1.0875003337860107 |
| Name | 1.0 |
| C.TIPO_ESPICULA_PRINCIPAL_ESTILO_TILOSTILO | 0.6074188947677612 |
But the model output shows:
Variable Relative Importance Scaled Importance Percentage
J.TIPO_DE_ESFERASTER 6.545000 1.000000 0.370537
P.PERFORANTE 5.506167 0.841278 0.311725
F.TIPO_DE_EXOSTILO 3.671128 0.560906 0.207836
C.TIPO_ESPICULA_PRINCIPAL_ESTRONGILOXA 1.087500 0.166157 0.061568
C.TIPO_ESPICULA_PRINCIPAL_ESTILO_TILOSTILO 0.607419 0.092807 0.034388
Note that "Name" attribute is once (falsely) listed among important attributes and once not.
Tagged:
0
Comments