Class JunkDetector.FeatureComponents

java.lang.Object
org.apache.tika.ml.junkdetect.JunkDetector.FeatureComponents
Enclosing class:
JunkDetector

public static final class JunkDetector.FeatureComponents extends Object
Per-feature z-score breakdown returned by JunkDetector.scoreWithFeatureComponents(String). z1-z3 are byte-count- weighted aggregates across script-homogeneous chunks; z4-z6 are single document-level values.

z5 (letter-adjacent-to-mark ratio) and z6 (replacement-character ratio) are document-level features included alongside z1–z4 in the per-script LR.

classifierWeights is the per-script linear classifier weight vector {w1, ..., wN, bias} for the dominant script — useful for hybrid models that recompute the logit after substituting one z-score with an externally-computed value.

  • Field Details

    • z1

      public final float z1
    • z2

      public final float z2
    • z3

      public final float z3
    • z4

      public final float z4
    • z5

      public final float z5
    • z6

      public final float z6
    • z7

      public final float z7
    • z8

      public final float z8
    • z9

      public final float z9
    • logit

      public final float logit
    • dominantScript

      public final String dominantScript
    • classifierWeights

      public final float[] classifierWeights
    • totalBytes

      public final int totalBytes