Warning: Rank deficient, rank = 4 tol = 1.7132e-003.

Discussion in 'MATLAB' started by Santi CH., May 2, 2004.

  1. Santi CH.

    Santi CH. Guest

    Hello

    I got a problem when running a simulation. The results came out, but
    it was a little deviated from what I expected. The warning messages
    are:

    Warning: Rank deficient, rank = 4 tol = 1.7132e-003.
    Warning: Rank deficient, rank = 4 tol = 1.7486e-003.
    Warning: Rank deficient, rank = 4 tol = 1.7913e-003.
    Warning: Rank deficient, rank = 4 tol = 1.8311e-003.
    Warning: Rank deficient, rank = 4 tol = 1.9166e-003.
    Warning: Rank deficient, rank = 4 tol = 1.9881e-003.

    I cant find the interpretation from the documents supplied with the
    program.

    Can anybody give me a hint or suggestion about this?

    1. What do this messages want to warn me?

    2. Do such messages indicate any potential errors in the results?

    3. Where can I find hints to solve the problem?

    Thank you in advance.
     
    Santi CH., May 2, 2004
    #1
    1. Advertisements

  2. Santi CH.

    Greg Heath Guest

    You can get warnings like that when radial basis functions are too fat
    and close together. Although many radial basis functions are generated,
    only 4 are associated with nonzero output weights. Furthermore,
    those 4 weights may have ridiculously high values so that an output
    with a small or moderate value is created by subtracting large numbers.
    Inappropriate input parameters. Vary them one at a time to try to
    understand what is going on.
    Potential numerical instability and poor generalization.
    Try the MATLAB newsgroup.

    Hope this helps.

    Greg
     
    Greg Heath, May 3, 2004
    #2
    1. Advertisements

  3. Santi CH.

    Santi CH> Guest

    Thank u very much Greg!

    I will try to get it back on the way.

    All the best,
     
    Santi CH>, May 4, 2004
    #3
  4. Santi CH.

    Santi CH. Guest

    Hi

    I finally can solve the problem. I normalized the matrix by dividing
    with a certain value before regression.

    This might solve the scaling problem i think.

    Thanks for everything
     
    Santi CH., May 4, 2004
    #4
  5. Santi CH.

    Vava Guest

    Vava, Oct 6, 2004
    #5
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.