Author Topic: Question about the Experimental Algorithm  (Read 5183 times)

Lupine

  • Guest
Question about the Experimental Algorithm
« on: September 24, 2009, 23:14:48 »
Is its results supposed to be more accurate? Like if %Content is 50% between two songs and experimental says 75%, should I listen to experimental more on closeness?

Also, on regular content similarity...at 0% I got a pair of two songs at about 65% similarity. When I did the content at 60% it was reduced to 30%. Shouldn't the similarity always be the same?

Thank you for your time

Lupine

  • Guest
Question about the Experimental Algorithm
« Reply #1 on: September 25, 2009, 13:42:57 »
Ok, I figured out %Content and it's sensitivity. From what I gather, if I set it to 0, I'll get a comparison...if I raise the sensitivity to over the %content, it will disappear since it is no longer close anymore. Using that, I guess I can tell how close a song is to its maximum sensitivity. More or less.

So using that if I have a song that's 50.2% similar to another song at 0% senitivity, then it will only be ~0.2% at 50%...so that means it's near 0 to comparing it to about even 50% closeness, right

Also having reread your post in someone else thread who asked a similar question...experimental is supposed to be more correct in closeness. Did I get it all right?

Admin

  • Administrator
  • Hero Member
  • *****
  • Posts: 624
    • View Profile
    • http://www.smilarityapp.com
Question about the Experimental Algorithm
« Reply #2 on: September 25, 2009, 16:56:51 »
Experemental algorithm will be changed in new versions for more robust and correct mechanism, it outperform old in every position, if all goes right it change current primary algorithm, and experemental algorithm will be deleted from program.

Scoring algorithm: (score-sensivity)/(1.0-sensivity), where score from 0 to 1.
That gives different scores for different sensivity.