Hi:
We are extracting feature importance in Scala using
model.nativeBooster. getScore("", "gain").
We are getting values that are above 1, more like order of 10s. First thought was they are percents, but closer look shows they all don’t add up to 100. So, how do we normalize these values between 0 and 1. Any thoughts, what could be going wrong?
scala> val featureScore = xgBoost.nativeBooster. getScore("", "gain")
featureScore: Map[String, Double] = Map(f25 -> 6.2915058125000005, f35 -> 7.848035394999999, f71 -> 14.902912531739128, f47 -> 7.04741025,
f87 -> 7.371472058135137, f20 -> 6.455246919999999, f60 -> 15.448602612148147, f19 -> 8.927439187142856, f36 -> 10.334689032222222, f76 -> 79.90250545497936, f7 -> 18.76505807365, f109 -> 2.421138782793182,
f95 -> 7.633686332857143. f72 -> 15.24958419714286. f81 -> 11.448275180000001, f68 -> 9.057547958580646, f110 -> 15.794339564000001, f92 -> 6.685504353333335, f6 -> 12.195955762782608,
f31 -> 12.0740433, f61 -> 10.12438512909091, f47 -> 6.750091075, f1 -> 22.412323225310345, f13 -> 10.53733687083334, f65 -> 25.554941611417973, f18 -> 8.68480279066667, f70 -> 9.97026484975,
f79 -> 20.02885708794118, f86 -> 5.793746619172414, f29 -> 3.15639432333333.