Difference between revisions of "Poissonscale"

From Eigenvector Research Documentation Wiki
Jump to navigation Jump to search
imported>Jeremy
(Created page with "===Purpose=== Perform Poisson scaling with scaling offset. ===Synopsis=== :[xs,sc] = poissonscale(x,''options''); %calibrate scaling :xs = poissonscale(x,sc); ...")
 
imported>Jeremy
(Options)
 
(2 intermediate revisions by the same user not shown)
Line 10: Line 10:
 
===Description===
 
===Description===
  
Scales each variable by its square root mean value. When no scale values
+
Scales each variable by its square root mean value. When no scale values are passed, a calibration is performed in which the square root mean values are calculated for each variable and then these are applied to the input data. If previously calculated scales are passed, these are simply applied to the data. An optional options structure allows setting of the offset which is added to each mean to avoid over-scaling when a variable has a near-zero mean.
are passed, a calibration is performed in which the square root mean
 
values are calculated for each variable and then these are applied to the
 
input data. If previously calculated scales are passed, these are simply
 
applied to the data. An optional options structure allows setting of the
 
offset which is added to each mean to avoid over-scaling when a variable
 
has a near-zero mean.
 
  
 +
In general, the larger the offset, the less influence the scaling will have. As the offset gets smaller, small values will be more up-scaled and larger values more down-scaled. At an offset of zero, artifacts from near-zero values can become a distraction so some small amount of offset (1-5%) is often recommended.
 +
 +
When used in [[Model_Building:_Preprocessing_Methods|Preprocessing Window]], the offset option is available through the settings interface.
  
 
====Inputs====
 
====Inputs====
Line 38: Line 35:
  
 
* '''offset''': [ 3] percent of the maximum mean value to be used as an offset on all scales. Avoids division by near-zero means.
 
* '''offset''': [ 3] percent of the maximum mean value to be used as an offset on all scales. Avoids division by near-zero means.
 +
 +
* '''mode''': [ 1 ] dimension of data on which to calculate the mean value for the scaling. 1 = mean over rows (to scale variables); 2 = mean over columns (to scale samples).
  
 
===See Also===
 
===See Also===
  
 
[[auto]], [[preprocess]], [[rescale]], [[scale]]
 
[[auto]], [[preprocess]], [[rescale]], [[scale]]

Latest revision as of 13:02, 9 June 2014

Purpose

Perform Poisson scaling with scaling offset.

Synopsis

[xs,sc] = poissonscale(x,options); %calibrate scaling
xs = poissonscale(x,sc); %apply previous scaling

Description

Scales each variable by its square root mean value. When no scale values are passed, a calibration is performed in which the square root mean values are calculated for each variable and then these are applied to the input data. If previously calculated scales are passed, these are simply applied to the data. An optional options structure allows setting of the offset which is added to each mean to avoid over-scaling when a variable has a near-zero mean.

In general, the larger the offset, the less influence the scaling will have. As the offset gets smaller, small values will be more up-scaled and larger values more down-scaled. At an offset of zero, artifacts from near-zero values can become a distraction so some small amount of offset (1-5%) is often recommended.

When used in Preprocessing Window, the offset option is available through the settings interface.

Inputs

  • x = Data to be scaled (double or DataSet object).

Optional Inputs

  • sc = Vector of previously-calculated scales. Must be equal in length to the number of included x-block variables.
  • options = Options structure. See Options section below.

Outputs

  • xs = Scaled data.
  • sc = Vector of scales calculated for given data.

Options

options = a structure array with the following fields:

  • offset: [ 3] percent of the maximum mean value to be used as an offset on all scales. Avoids division by near-zero means.
  • mode: [ 1 ] dimension of data on which to calculate the mean value for the scaling. 1 = mean over rows (to scale variables); 2 = mean over columns (to scale samples).

See Also

auto, preprocess, rescale, scale