Martin Walter
Martin Walter
It seems this `--match` argument is broken, I tried all solution s I can find in other issues but still does not work.
@acartro I have a PR, now there is just one question. When `refit is False`, the last window will be used to fit the best forecaster instead of all data....
probably the "cleanest" solution would be to move the last window 3 points in the future in above example with `fh=[1,2,3]`. With `update` the window size would increase by 3...
Did it now with `update`, pls review
Will come back to this PR next week 😊
> that's attaching "too much responsibility" I think that should not be an argument, we should always design from a user perspective and not how convenient it is for us...
it does not raise error but I get weird predictions: ```python pipe = TransformedTargetForecaster(steps=[ ("boxcox", BoxCoxTransformer()), ("difference", Differencer()), ("forecaster", NaiveForecaster())]) pipe.fit(y) pipe.predict(fh=3) ``` ``` >>> 1961-01 477.793697 >>> 1961-02 527.659894...
I think actually that forecaster should live in forecasting module, but for now as it is experimental it can stay where it is probably
My name is fine for multiple estimator types