llc-r1 Locally Linear Coordination, release 1 ============================================== Parameters ========== NN Number of data points. nn Number of nearest neighbours. DD Dimension of data points. dd Dimension of embedding. KK Number of components in MFA. FF Dimension of latent variables in MFA. numiter Maximum number of iterations of EM for MFA. tol Termination tolerance of EM for MFA. zerotol Tolerance before setting responsibilities to 0. nzr Number of non zero entries in RR. nza Number of non zero entries in AA. nzb Number of non zero entries in BB. Variables ========= (data and embedding) XX DD x NN Data vectors. YY dd x NN Embeddings of data. (mfa) Pi KK x 1 Mixing proportions of components in MFA. LX DD x FF x KK Factor loadings to XX. MX DD x KK Means of factor analyzers to XX. PX DD x 1 Noise covariance of XX. (embedding representation) LY dd x FF x KK Factor loadings to YY. MY dd x KK Means of factor analyzers to YY. (data representation by mfa) RR KK x NN Responsibilities of components of MFA. ZZ FF x NN x KK Mean of posteriors over latent variables of MFA. UU NN x KK(FF+1) Latent representation of data by MFA. (data representation by neighbour reconstruction) WW nn x NN sparse Matrix of convex reconstruct weights. MM NN x NN sparse Quadratic cost matrix of LLE. Directories =========== demo Demos on freyfaces, mnist, trefoil, twinpeaks, usps. mfa Mixtures of factor analyzers. curves Simple curves and running LLC on them. mnist Running LLC on MNIST data set. usps Running LLC on USPS data set. freyfaces Running LLC on Brendan Frey's faces. joshfaces Running LLC on Josh's faces dataset (not done). backup/* Some previous versions of code. backup/lle Sam's LLE code. backup/gcview Sam's routines for viewing global coordinates. backup/vbmfa Matt's Variational Bayes MFA. backup/zoubin_mfa Zoubin's MFA code. M-files =======