/ Data / Transform

Transform measured values to remove scale effect, extract information, or re-calibrate

Once data are available as matrices of numbers where no data are missing anymore, data transformations are commonly applied. Some common transformations are reported in the following table.

Name Description
Center To bring the expected mean values back to zero.
Scale To remove the effect of the scale.
Recalibrate To emphasize, e.g., small or large values.
Normalize To center and bring values onto comparable scales.
Indicator To extract a single level or value from a variable.
Quantile To disregard value differences but keep order.
Tf-idf Each wordโ€™s contribution is weighted as a function of the term frequency (tf) and its inverse frequency in all text documents (idf)

The transformed numbers compose the set of predictor (or independent) variables, which statistical methods rely upon to build statistical models.