Ch. 10: Linear Discriminant Analysis (LDA) based on slides from Duncan Fyfe Gillies Carlos Thomaz authored the original version of these slides Modified by Longin Jan Latecki Temple University latecki@temple.edu Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 1 The story so far: Let us start with a data set which we can write as a matrix: X = x1,1 x2,1 x3,1 x1,2 x2,2 xn,1 x1,N x2,N xn,N Each column is one data point, each row is a variable, but take care sometimes the transpose is used Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 2 The mean adjusted data matrix We form the mean adjusted data matrix by subtracting the mean of each variable U = x1,1 - m1 x1,2 - m1 x2,1 - m2 x2,2 - m2 x3,1 - m3 x1,N - m1 x2,N - m2 xn,1 - mn xn,N - mn mi is the mean of the data items in row i Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 3 Covariance Matrix The covariance matrix can be formed from the product: S = S = (1/(N-1)) U UT Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 4 Alternative notation for Covariance Covariance is also expressed in the following way: SS= 1 ( N - 1) N j =1 ( x j - x )( x j- x ) T 1xn row nx1 column Sum over the data points nxn matrix Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 5 Projection A projection is a transformation of data points from one axis system to another. 1 2 X2 (xi-m1 xi-m m X1 xi Finding the mean adjusted data matrix is equivalent to moving the origin to the center of the data. Projection is then carried out by a dot product Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 6 Projection Matrix A full projection is defined by a matrix in which each column is a vector defining the direction of one of the new axes. F = [ 1, 2, 3, . . . m ] Each basis vector has the dimension of the original data space. The projection of data point xi is: yi = (xi - m)T F Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 7 Projecting every point The projection of the data in mean adjusted form can be written: UT F = ( FT U )T Projection of the covariance matrix is FT S F = FT (1/(N-1)) U UT F = (1/(N-1)) (FT U)(UT F ) which is the covariance matrix of the projected points Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 8 Orthogonal and Orthonormal For most practical cases we expect the projection to be orthogonal (all the new axes are at right angles to each other) and orthonormal (all the basis vectors defining the axes are unit length) thus: FT F = I Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 9 PCA The PCA projection is the one that diagonalises the covariance matrix. That is it transforms the points such that they are independent of each other. FT S F = L We will look at another projection today called the LDA. Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 10 Introduction Ronald A. Fisher, 1890-1962 “The elaborate mechanism built on the theory of infinitely large samples is not accurate enough for simple laboratory data. Only by systematically tackling small sample problems on their merits does it seem possible to apply accurate tests to practical data.” 1936 Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 11 Introduction What is LDA? Linear Discriminant Analysis, or simply LDA, is a well-known feature extraction technique that has been used successfully in many statistical pattern recognition problems. LDA is often called Fisher Discriminant Analysis (FDA). Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 12 Motivation The primary purpose of LDA is to separate samples of distinct groups by transforming then to a space which maximises their between-class separability while minimising their within-class variability. It assumes implicitly that the true covariance matrices of each class are equal because the same within-class scatter matrix is used for all the classes considered. Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 13 Geometric Idea x2 u 1 2 PCA: (1,2) x2 LDA: u x1 Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 14 x1 Method Let the between-class scatter matrix Sb be defined as Sb = g N (x i i - x )( xi - x ) T i =1 and the within-class scatter matrix Sw be defined as Sw = g i =1 ( N i - 1)Si = g Ni ( xi , j - xi )( xi , j - xi )T i =1 j =1 where xi,j is an n-dimensional data point j from class pi, Ni is the number of training examples from class pi, and g is the total number of classes or groups. Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 15 Nota Bene A scatter matrix is un-normalised, using the data matrix formulation: S = U UT is a scatter matrix but Scov = U UT /(N-1) is the corresponding covariance matrix Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 16 Method The Sw matrix is essentially made up from the pooled estimate of the covariance matrix: 1 Sp = N-g g ( N i - 1)Si = ( N1 - 1)S1 ( N 2 - 1)S 2 ( N g - 1)S g i =1 N-g Sw = (N-g) Sp Since each Si has rank Ni -1 its rank can be at most N-g Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 17 Method (cont.) The sample mean, sample covariance, and grand mean vector are given respectively by: 1 xi = Ni Ni x j =1 i, j Ni 1 T Si = ( x x )( x x ) i, j i i, j i ( Ni - 1) j =1 1 x= N g 1 N i xi = N i =1 g Ni x i, j i =1 j =1 Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 18 N = N1 N2 N g Method (cont.) The main objective of LDA is to find a projection matrix Plda that maximises the ratio of the determinant of Sb to the determinant of Sw (Fisher’s criterion), that is T Plda = arg max P P Sb P PT S w P Intuition (cont) So Fisher’s criterion tries to find the projection that: Maximises the variance of the class means Minimises the variance of the individual classes Poor projection axis for separating the classes Best (LDA) projection axis for separating the classes Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 20 Good projection Bad projection Intuition The determinant of the co-variance matrix tells us how much variance a class has. Consider the co-variance matrix in the PCA (diagonal) projection - the determinant is just the product of the diagonal elements which are the individual variable variances. The determinant has the same value under any orthonormal projection. Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 22 Method (cont.) It has been shown that Plda is in fact the solution of the following eigensystem problem: Sb P - S w PL = 0 Multiplying both sides by the inverse of Sw S w-1Sb P - S w-1S w PL = 0 S w-1Sb P - PL = 0 (S w-1Sb ) P = PL Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 23 Standard LDA If Sw is a non-singular matrix then the Fisher’s criterion is maximised when the projection matrix Plda is composed of the eigenvectors of -1 w b S S with at most (g-1) nonzero corresponding eigenvalues. (since there are only g points to estimate Sb) Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 24 Classification Using LDA The LDA is an axis projection. Once the projection is found all the data points can be transformed to the new axis system along with the class means and covariances. Allocation of a new point to a class can be done using a distance measure such as the Mahalanobis distance. Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 25 LDA versus PCA LDA seeks directions that are efficient for discriminating data whereas PCA seeks directions that are efficient for representing data. The directions that are discarded by PCA might be exactly the directions that are necessary for distinguishing between groups. Intelligent Data Analysis and Probabilistic Inference Lecture 17 Slide No 26