SDSCALE Scaling data Train scaling transformation from data: PS=SDSCALE(DATA,METHOD) Define scaling by multiplication and offset vectors: PS=SDSCALE(MUL,OFFSET) Apply non-linear scaling to a pipeline: P2=SDSCALE(P,'EXP') INPUT DATA Data set METHOD Scaling method (default: 'standardization') MUL Multiplication vector OFFSET Offset vector OUTPUT PS Scaling pipeline DESCRIPTION SDSCALE trains a scaling pipeline from DATA or crates manually-defined scaling transformation from multiplication and offset vectors. Scaling methods: 'standardization' - shift data to zero mean and scale to unit variance (default, synonym: 'variance') 'centering' - mean centering shifting data to zero mean 'range' - scale data into 0-1 interval using min and max data values 'robust' - scale the bulk of data into 0-1 range using lower and upper percentiles. This is robust to outliers. Percentile may be set with an additional parameter (default: 0.05) 'reg',R - for 'variance' method: add regularization parameter R to standard deviation to avoid NaNs on sparse data. Example: ps=sdscale(small_data,'reg',eps) Non-linear scaling methods: 'exp',A - exp(x*A) (default A=1.0) 'log',A - log(x+A) (default A=1.0) 'log2',A - log(abs(x)+A) (default A=1.0)
sdscale
is referenced in examples: