SVDer

SVDer performs singular value decomposition of the matrix of expression values in a given PCL file (which should contain no missing values; see KNNImputer). The PCL file can then be transformed in SVD space using a number of algorithms, e.g. up- or down-weighting low-variance singular values or continuously transforming the weights of each value.

Usage

Basic Usage

 SVDer -i <input.pcl> -o <output.pcl> -r <fraction>

Decompose the input matrix input.pcl, retain only the minimum number of singular values necessary to account for fraction of the total variance, set the rest to zero, and reproject the resulting matrix in output.pcl.

 SVDer -i <input.pcl> -o <output.pcl> -b

Decompose the input matrix input.pcl, set all singular values to the mean (thus preserving the total variance), and reproject the resulting matrix in output.pcl.

 SVDer -i <input.pcl> -o <output.pcl> -t <parameter>

Decompose the input matrix input.pcl, transform the singular values proportionally to the parameter power, and reproject the resulting matrix in output.pcl.

Detailed Usage

package "SVDer"
version "1.0"
purpose "Transforms PCL data through Support Vector Decomposition"

section "Main"
option  "input"         i   "Input PCL file"
                            string  typestr="filename"
option  "output"        o   "Output PCL file"
                            string  typestr="filename"
option  "umatrix"       u   "Output U matrix PCL file"
                            string  typestr="filename"

section "SVD"
option  "reprojection"  r   "Fraction of variance to reproject"
                            double  default="1"
option  "transform"     t   "Transformation function parameter"
                            double  default="1"
option  "signal_balance"    b   "Equally weight all SVs"
                            flag    off

section "Optional"
option  "skip"          s   "Columns to skip in input PCL"
                            int default="2"
option  "verbosity"     v   "Message verbosity"
                            int default="5"

Flag	Default	Type	Description
-i	stdin	PCL file	Input PCL file to be transformed; must not contain any missing values.
-o	stdout	PCL file	Output PCL file containing the transformed version of -i.
-u	None	PCL file	If given, output PCL file containing the U matrix (basis vectors) of the decomposition of -i.
-r	1	Double (fraction)	Fraction of variance to be reprojected; when less than 1, proportionally many singular values are set to zero before reprojecting the output matrix.
-t	1	Double	Power by which singular values are transformed before reprojection. For singular values {s1, ..., sn} S = sum(si), ti = (si/S)^-t, and T = sum(ti), each si is replaced by ti/sum(ti).
-b	Off	Flag	If given, set all singular values to their average value before reprojection; otherwise, transform them according to -t.
-s	2	Integer	Number of columns to skip between the initial ID column and the first experimental (data) column in the input PCL.