post.processing: Post processing meta-features values

In mfe: Meta-Feature Extractor

Description

Post-processing alternatives to deal with multiples meta-features values. This method is used by the meta-features characterization methods to summarize the obtained values.

Usage

post.processing(measure, summary = c("mean", "sd"), multiple = TRUE, ...)

Arguments

measure	A list with the meta-features values.
summary	The functions to post processing the data. See the details to more information. Default: c("mean", "sd")
multiple	A logical value defining if the measure should return multiple values. (Default: TRUE)
...	Extra values used to the functions of summarization.

Details

The post processing functions are used to summarize the meta-features. They are organized into three groups: non-aggregated, descriptive statistic and distribution. Currently, the hypothesis testing post processing are not supported.

In practice, there are no difference among the types, so that more than one type and functions can be combined. Usually, these function are used to summarize a set of values for each meta-features. For instance, a measure computed for each attribute can be summarized using the "mean" and/or "sd". Necessarily, a single value always use the "non.aggregated" function.

In addition to the native functions available in R, the following functions can be used:

"histogram"

Computes a histogram of the given data value. The extra parameters 'bins' can be used to define the number of values to be returned. The parameters 'max' and 'min' are used to define the range of the data. The default value for these parameters are respectively 10, min(x) and max(x).

"kurtosis"

See kurtosis

"max"

See max

"mean"

See mean

"median"

See median

"min"

See min

"quantiles"

See quantile

"sd"

See sd

"skewness"

See skewness

"var"

See var

"non.aggregated"

Returns the original value(s) of the meta-feature.

These functions are not restrictive, thus another functions can be applied as post-processing summarization function.

Value

A list with the post-processed meta-features

References

Fabio Pinto, Carlos Soares, and Joao Mendes-Moreira. Towards Automatic Generation of Metafeatures. In 20th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD), pages 215 - 226, 2016.

Examples

post.processing(runif(15))
post.processing(runif(15), c("min", "max"))
post.processing(runif(15), c("quantiles", "skewness"))
post.processing(runif(15), "histogram", bins=5, min=0, max=1)

Example output

     mean        sd 
0.5834315 0.2723582 
       min        max 
0.03067458 0.98869939 
  quantiles.0%  quantiles.25%  quantiles.50%  quantiles.75% quantiles.100% 
     0.1170565      0.2001647      0.5657831      0.6898982      0.8586654 
      skewness 
    -0.2255150 
histogram1 histogram2 histogram3 histogram4 histogram5 
0.06666667 0.46666667 0.20000000 0.13333333 0.13333333 

mfe documentation built on July 1, 2020, 10:46 p.m.