You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
One addon possibility that I was wondering:
In case of time series, de-trending is a common operation, done by taking the first differential of the data eg. with pd.DataFrame.diff(1). The problem with this is, that the initial value gets dropped and there is no easy way to "back transform".
Here a "fittable", Scikit like transformer could come in handy.
Yes, definitely! Though I have to say no pdpipe stage at the moment has an inverse_transform method, so you still won't have invertible pipelines... :|
Yep, the sklearn ones can definitely be made invertible.
The NLTK ones for sure don't. For example, if you drop rare tokens or stem words, you have no way to go back, as these are transformations that map many different inputs into the same output (e.g. "grabbing" and "grabbed" to "grab").
Hi,
Pdpipe is marvelous, very nice tool!
One addon possibility that I was wondering:
In case of time series, de-trending is a common operation, done by taking the first differential of the data eg. with
pd.DataFrame.diff(1)
. The problem with this is, that the initial value gets dropped and there is no easy way to "back transform".Here a "fittable", Scikit like transformer could come in handy.
I have sketched such a thing for myself here: https://gist.github.com/solalatus/9a3fc5330e7c0cd83e61094db75d2dc3
Can this be interesting as an addition?
Many thanks!
The text was updated successfully, but these errors were encountered: