U Kv f!@s,dZddlZdddZd ddZddgZdS) zPrincipal Component Analysis Created on Tue Sep 29 20:11:23 2009 Author: josef-pktd TODO : add class for better reuse of results NTc Cst|}|r|d}nt|jd}||8}tj|dd}tj|\}}t|} | ddd} |dd| f}|| }|dkr||jdkr|ddd|f}|d|}|r|t |}t ||} t | |j |} | | ||fS)aprincipal components with eigenvector decomposition similar to princomp in matlab Parameters ---------- data : ndarray, 2d data with observations by rows and variables in columns keepdim : int number of eigenvectors to keep if keepdim is zero, then all eigenvectors are included normalize : bool if true, then eigenvectors are normalized by sqrt of eigenvalues demean : bool if true, then the column mean is subtracted from the data Returns ------- xreduced : ndarray, 2d, (nobs, nvars) projection of the data x on the kept eigenvectors factors : ndarray, 2d, (nobs, nfactors) factor matrix, given by np.dot(x, evecs) evals : ndarray, 2d, (nobs, nfactors) eigenvalues evecs : ndarray, 2d, (nobs, nfactors) eigenvectors, normalized if normalize is true Notes ----- See Also -------- pcasvd : principal component analysis using svd r)ZrowvarN) nparraymeanzerosshapeZcovlinalgZeigZargsortsqrtdotT) datakeepdim normalizedemeanxmZxcovevalsZevecsindicesfactorsxreducedrG/tmp/pip-unpacked-wheel-2v6byqio/statsmodels/sandbox/tools/tools_pca.pypcas&#     rc Cs|j\}}t|}|r$|d}nd}||8}tjj|jdd\}}} t|j|jj} |rt| ddd|f|ddd|fj|} n|} |}d|f|d|jdd} | | ddd|f| d||ddd|ffS)aprincipal components with svd Parameters ---------- data : ndarray, 2d data with observations by rows and variables in columns keepdim : int number of eigenvectors to keep if keepdim is zero, then all eigenvectors are included demean : bool if true, then the column mean is subtracted from the data Returns ------- xreduced : ndarray, 2d, (nobs, nvars) projection of the data x on the kept eigenvectors factors : ndarray, 2d, (nobs, nfactors) factor matrix, given by np.dot(x, evecs) evals : ndarray, 2d, (nobs, nfactors) eigenvalues evecs : ndarray, 2d, (nobs, nfactors) eigenvectors, normalized if normalize is true See Also -------- pca : principal component analysis using eigenvector decomposition Notes ----- This does not have yet the normalize option of pca. rr)Z full_matricesNz print reassigning keepdim to max)rrrrr Zsvdr r ) r rrZnobsZnvarsrrUsvrrrrrrpcasvdWs!   4r)rrT)rT)__doc__Znumpyrrr__all__rrrrs  I =