Mixed Euclidean Distance (Distance Metrics)

PinguiculaPinguicula Member Posts: 12 Contributor II
Hi,

Rapidminer offers  "mixed Euclidean Distance" (MED) as distance measure. In contrast to the other distance measures
I can not find MED as technical term in the text books and internet ressources available to me. I' d like to ask where I can find a reference which explains the computaion of this metric. Or does it refer to Euclidean Distance for datasets with mixed discrete and continuous variables?

Best

Norbert

Answers

  • steffensteffen Member Posts: 347  Guru
    Yes !

    Quote the Java class:
    Euclidean distance for numerical and nominal values. For nomimal values, a distance of one is accounted if both values are not the same.
  • kostanskikostanski Member Posts: 1 Contributor I
    It is worth quota the full comment in the code of this class:

    Euclidean distance for numerical and nominal values. For nomimal values, a distance of one is accounted if both values are not the same. Note: In most cases, you must normalize the numerical values, to obtain sound results.

    So I understand, that it DOES NOT calculate normalized distance by default. If I am wrong, let me know.
  • MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869   Unicorn
    You are right.
    Almost all distance measures do not normalize the data automatically.

    Best regards,
    Marius
Sign In or Register to comment.