Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 1

    Uploaded by

    VIKRAM KUMAR
    35 views2 pages
    1. The document provides methods for quickly calculating a new discriminant (S-1 new) when adding or removing a variable from subset selection of normally distributed classes. It explains that the inverse of a partitioned symmetric matrix can be calculated from the inverse of the top left partition and the new variable covariances, avoiding recalculating the entire inverse. 2. The document discusses implementing PCA on the Optdigits dataset from UCI and reconstructing digit images using various numbers of eigenvectors. It calculates the reconstruction error defined by equation 6.12 from the machine learning textbook. Code is provided in an external file.

    Original Description:

    DVCVCVCB

    Copyright

    © © All Rights Reserved

    Available Formats

    DOCX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document
    1. The document provides methods for quickly calculating a new discriminant (S-1 new) when adding or removing a variable from subset selection of normally distributed classes. It explains that the inverse of a partitioned symmetric matrix can be calculated from the inverse of the top left partition and the new variable covariances, avoiding recalculating the entire inverse. 2. The document discusses implementing PCA on the Optdigits dataset from UCI and reconstructing digit images using various numbers of eigenvectors. It calculates the reconstruction error defined by equation 6.12 from the machine learning textbook. Code is provided in an external file.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Flag for inappropriate content
    Download as docx, pdf, or txt
    35 views2 pages

    1

    Uploaded by

    VIKRAM KUMAR
    1. The document provides methods for quickly calculating a new discriminant (S-1 new) when adding or removing a variable from subset selection of normally distributed classes. It explains that the inverse of a partitioned symmetric matrix can be calculated from the inverse of the top left partition and the new variable covariances, avoiding recalculating the entire inverse. 2. The document discusses implementing PCA on the Optdigits dataset from UCI and reconstructing digit images using various numbers of eigenvectors. It calculates the reconstruction error defined by equation 6.12 from the machine learning textbook. Code is provided in an external file.

    Copyright:

    © All Rights Reserved
    Download as DOCX, PDF, TXT or read online from Scribd
    Flag for inappropriate content
    Download as docx, pdf, or txt
    of 2

    1.

    Assuming that the classes are normally distributed, in subset selection, when one
    variable is added or removed, how can the new discriminant be calculated quickly? For
    example, how can the new S−1 new be calculated from S−1 old?
    ANSWER

    When we add a new variable, we are adding a row and column to the covariance matrix. It
    can be shown that if we partition a symmetric nonsingular matrix A into (Rencher, 1995; p.
    28)

    the inverse is given by

    where b = a22 − a T 12A −1 11 a12. So given A11 and A −1 11 , we can easily calculate A−1
    when a12 and a22 (covariances of the new variable with the existing variables, and its
    variance) are added.

    Another possibility is to break down the Mahalanobis distance calculation (if we have the
    previous values stored). Let A be partitioned as (T. Cormen, C. Leiserson, R. Rivest, C. Stein.
    2001. Introduction to Algorithms, The MIT Press. 2nd edition, p. 761)

    2. Using Optdigits from the UCI repository, implement PCA. For various number of
    eigenvectors, reconstruct the digit images and calculate the reconstruction error
    (equation 6.12).

    ANSWER
    The Matlab code is given in ex6_2.m (which also generates figure 6.3 in the book). The
    contour plot of an example ‘6’ and its reconstruction is given in figure 6.1. The original 64
    dimensional data and its reconstruction are as follows:

    You might also like