Add option to drop rolling window sequences
csala opened this issue · 0 comments
csala commented
Add a new argument to the custom.timeseries_preprocessing.rolling_window_sequences
primitive to be able to selectively drop some of the generated sequences.
This should be new produce
argument called drop
which indicates which values in the timeseries sequence are invalid and have to be dropped. This argument would be required=False
, following this specification.
The input can be of several types:
numpy.ndarray
(or array-like): 1d array-like object of boolean (or binary integer) values that indicate which values are invalid. This references both to X and y indexes.None
,str
,int
,float
orbool
: Any window that contains this value withinX
ory
will be dropped.
Additionally, a second argument called drop_windows
should be added to disable the dropping altogether. This should be set as a fixed hyperparameter with default value False
to keep the current behavior by default.