revise_forecasts() method raises an attribute error. AttributeError: 'Series' object has no attribute 'yhat' about scikit-hts

[carlomazzaferro]

This was due to bad assumptions on the format of the forecasts dict. It was assumed that it is a DataFrame with a single columns, named yhat.

This PR address it. Feel free to take a look, the tests should make it clear what the PR is achieving.

[aakashparsi]

Thanks for the update @carlomazzaferro.
I just want to clarify one thing here. So, according to the updated code, To revise the forecasts we have to pass the DataFrame with one column. Is that right?

[aakashparsi]

And, the same error pops up when using optimal_combination().
Here is the stack trace

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-152-a556a14422e8> in <module>
----> 1 optimal_combination(dict(pred[-4:]), sum_mat = htsmodel.sum_mat, method = 'OLS', mse = {} )

~/.conda/envs/lts/lib/python3.6/site-packages/hts/functions.py in optimal_combination(forecasts, sum_mat, method, mse)
    128 
    129     """
--> 130     hat_mat = y_hat_matrix(forecasts)
    131     transpose = np.transpose(sum_mat)
    132 

~/.conda/envs/lts/lib/python3.6/site-packages/hts/functions.py in y_hat_matrix(forecasts, keys)
     89         keys = forecasts.keys()
     90     first = list(forecasts.keys())[0]
---> 91     y_hat_mat = np.zeros([len(forecasts[first].yhat), 1])
     92     for key in keys:
     93         f1 = np.array(forecasts[key].yhat)

~/.conda/envs/lts/lib/python3.6/site-packages/pandas/core/generic.py in __getattr__(self, name)
   5137             if self._info_axis._can_hold_identifiers_and_holds_name(name):
   5138                 return self[name]
-> 5139             return object.__getattribute__(self, name)
   5140 
   5141     def __setattr__(self, name: str, value) -> None:

AttributeError: 'Series' object has no attribute 'yhat'

[carlomazzaferro]

You should not use optimal_combination directly. The new function that perform sanitization on the inputs should transform whatever you pass as your forecasts to a format that should work with the revised = revision.revise(forecasts=forecasts, mse=errors, nodes=nodes) (optimal_combination gets called by this method).

Now, you should be able to pass as forecasts a dict of strings to either numpy.array (as long as the array has ndim == 1), pandas.Series or pandas.DataFrame (as long as the df has only one column).

But I suggest you to wait till tomorrow, as I'm still working on some further fixes to the convenience methods. You can track the progress here: #68

[aakashparsi]

Hey @carlomazzaferro, your quick response is appreciated. I'll check out #68 for further updates.

[carlomazzaferro]

Feel free to bump to scikit-hts 0.5.6, see updated convenience docs https://scikit-hts.readthedocs.io/en/latest/hts.html#module-hts.convenience

Thanks for your patience!

[aakashparsi]

Sure @carlomazzaferro. Also, It would be very helpful If you throw some light on #66.

[carlomazzaferro]

Yes, that is next on my task list and will tackle that together with the related issue about fixing the transforms functionality. Will keep posted when I have a PR in the works

[aakashparsi]

Sure @carlomazzaferro. I Wholeheartedly appreciate all your efforts on this. Thank you so much.