An instance filter that assumes instances form time-series data and replaces attribute values in the current instance with the difference between the current value and the equivalent attribute attribute value of some previous (or future) instance. For instances where the time-shifted value is unknown either the instance may be dropped, or missing values used. Skips the class attribute if it is set.
Valid options are:
Specify list of columns to translate in time. First and
last are valid indexes. (default none)
Invert matching sense (i.e. calculate for all non-specified columns)
The number of instances forward to translate values
between. A negative number indicates taking values from
a past instance. (default -1)
For instances at the beginning or end of the dataset where
the translated values are not known, remove those instances
(default is to use missing values).