Results 1 to 3 of 3

Thread: Normalization

  1. #1

    Thread Starter
    Super Moderator Shaggy Hiker's Avatar
    Join Date
    Aug 2002
    Location
    Idaho
    Posts
    39,437

    Normalization

    I have a variety of variables that might be part of an equation. A look at the graph of these variables shows that they all have roughly similar distributions, but some have a range of a few thousand to a few hundred thousand (6K - 200K), while others range from perhaps -5 to 5, while others are 33.X, where X ranges from 0.6 to 0.9.

    In theory, for my purposes, these different ranges shouldn't matter all that much, but what I'm doing is highly speculative, so theory can be quite wrong. So, what I'm looking to do is shift them all such that they all have roughly the same range. A simple approach to that was to take the mean of the variable with the highest range (the one that varies from 6K to 200K), and multiply all the others by roughly that mean. It wasn't exactly that mean, as any that ranged from 10 to 30, might be multiplied by a tenth of that mean, while ones that were varying from 0 to 1 would be multiplied by the mean, but the idea was to exaggerate the variability in the variables with a smaller range.

    What I'm looking for is alternate transformations that would retain the variability within each variable, but scale it up to the magnitude of the largest of the variables. Any suggestions?
    My usual boring signature: Nothing

  2. #2
    PowerPoster
    Join Date
    Nov 2017
    Posts
    3,366

    Re: Normalization

    Assuming you want a transformation of, for example, your -5 to 5 range into an "equivalent" 6000 to 200000 range, then the "worked out math" would be something like this:

    -5*a+b=6000
    5*a+b=200000
    2*b=206000
    b=103000
    5*a+103000=200000
    a=19400

    So the transform from x where -5<=x<=5 would be:
    x' = 19400*x + 103000 (The ' just notates that it is a transformed value, not a derivative notation)

    The process to calculate the slope and intercept of the transform is just a generalized solution of two simultaneous equations with two unknowns, which I'm sure you can handle in code.

  3. #3

    Thread Starter
    Super Moderator Shaggy Hiker's Avatar
    Join Date
    Aug 2002
    Location
    Idaho
    Posts
    39,437

    Re: Normalization

    I had forgotten about this thread. I got distracted by a different aspect of the problem and moved away from the normalization. I did see this, and sat and pondered whether or not I really wanted to do that. I'm still not sure.
    My usual boring signature: Nothing

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  



Click Here to Expand Forum to Full Width