Decorrelation Projection Scalable to High Dimensional Data
Create auto-correlation matrix
Canonical correlation analysis
Estimate covariance matrix after applying transformation
Decorrelation projection
Multiply by diagonal matrix
Estimate covariance/correlation with low rank and shrinkage
Compute eclairs decomp of squared correlation matrix
Class eclairs
Estimate covariance/correlation with low rank and shrinkage
Class fastcca
Fast canonical correlation analysis
Get full covariance/correlation matrix from eclairs
Estimate shrinkage parameter by empirical Bayes
Get whitening matrix
Compute condition number
Fit linear model on each feature after decorrelating
Fit linear model after decorrelating
Evaluate the log determinant
Mahalanobis Distance
Multiply by eclairs matrix
Optimal Hard Threshold for Singular Values
Plot eclairs object
Evaluate quadratic form
Recompute eclairs after dropping features
Draw from multivariate normal and t distributions
Summarize correlation matrix
Singular value thresholding
Decorrelation projection + eclairs
Data whitening is a widely used preprocessing step to remove correlation structure since statistical models often assume independence. Here we use a probabilistic model of the observed data to apply a whitening transformation. This Gaussian Inverse Wishart Empirical Bayes model substantially reduces computational complexity, and regularizes the eigen-values of the sample covariance matrix to improve out-of-sample performance.
Useful links