sktime: [BUG] Auto-ETS parameter setting not working

Describe the bug Want to set auto=True, damped_trend=True, seasonal= None and bound={damping_bound:(0,0.5)} for Auto-ETS–>

To Reproduce import sktime import pandas as pd import numpy sa np import statsmodel def model_training(y_train):

    model_fcst = AutoETS(auto=True, additive_only=True, sp=12, trend='add', damped_trend=True,
                     allow_multiplicative_trend=False, bounds= {damping_bound:(0,0.5)}, error='add',
                      n_jobs=-1)
    model_fcst.fit(y_train)
    return model_fcst
import sktime
import pandas as pd
import numpy sa np
import statsmodel
def model_training(y_train):

        model_fcst = AutoETS(auto=True, additive_only=True, sp=12, trend='add', damped_trend=True,
                         allow_multiplicative_trend=False, bounds= {damping_bound:(0,0.5)}, error='add',
                          n_jobs=-1)
        model_fcst.fit(y_train)
        return model_fcst```

**Expected behavior**

I get the following error:
Arguments: (ValueError('Invalid key: smoothing_seasonal in bounds dictionary'),). It basically doesn't allow me to set auto=True and yet manually specify the damping_bounds range


**Versions**

sktime 0.13.2, python 3.9
Even tried on a different environment with sktime 0.22 and python 3.11

</details>

<!-- Thanks for contributing! -->

About this issue

  • Original URL
  • State: open
  • Created 10 months ago
  • Comments: 31 (1 by maintainers)

Most upvoted comments

I haven’t been an active contributor for a long time

But this might be the golden opportunity for a glorious return, @HYang1996! (you’re a former core dev so it’s easy to “reactivate”)

Currently what happens: In auto-ets if I set auto=True, it doesn’t acknowledge any other hyper-parameters I have set such as additive_only=True, trend=‘add’, damped_trend=True,seasonal=None, bounds, etc. It chooses the best model across all possibilities.

My expectation: In auto-ets, if i set auto =True, then it should acknowledge all hyper-parameters that i have set and choose best model acknowledging my specifications like trend=‘add’, seasonal=None, etc.

If this request is incorporated in the latest sktime release version, i can upgrade to it and find other ways to deal with library version conflicts or even kick it off in a separate docker container also.