0

I'm experimenting with ELKI (which is awesome, btw) and would like to try the weighted Euclidean distance function as a metric for the DBSCAN algorithm.

First of all, I don't know how it works, except for this.

I tried a few values for the parameter distance.weights. If I insert for example 3,1 in its field, this is what I get after I hit ENTER:

Screenshot

I'm still able to run the task but I see no difference in the results, no matter the values I insert.

Could anyone please provide a short explanation on how the function works and eventually what I'm doing wrong?

Thanks in advance.

BSheep
  • 1
  • 1
  • Since 1*x=x, a weight of 1 does not change anything. Weights are linear scaling factors. – Erich Schubert Jul 24 '17 at 20:59
  • Hi Erich! It was just to illustrate, but I admit it was a bad example. I edited and changed the values. Is it normal that I get 1.0,0.0 after hitting ENTER? – BSheep Jul 25 '17 at 17:17
  • If you enter "1,0" then yes. The default double formatter will format 1 as 1.0. Weight 0 = ignore that column. Depending on your data, the second column may simply have no influence, then changing the weight won't change the result. What data do you use? – Erich Schubert Jul 26 '17 at 12:24
  • Actually in the example above I entered 3,1. My data has two columns, x and y, representing coordinates in the Cartesian plane. – BSheep Jul 31 '17 at 17:23
  • It should not turn 3,1 into 1,0; there probably was a bug a year ago in the function converting this back to a string (I have a vague memory of a off-by-one bug in the rarely used double-list parameter class). But this does not happen for me anymore in the Git version, please use the latest code instead. – Erich Schubert Aug 01 '17 at 09:22

0 Answers0