Hi @fkiraly, here are my suggested changes, let me know what you think!

I think this will mess up the tags, since assignments of the _tags class variable will override each other and not give you the set union of dictionary elements.

I would really really like to go with the current, working workaround, and we make changes like this later on?

In particular, the mixins need to be completely removed in a second step, I was just going for an intermediate state where we bundle the logic in the base class.

@fkiraly I prefer this solution, this way, when we remove the mixins, we only have to copy the tags to the concrete forecasters, plus we don't change the constructor. Note that tags are not overridden, that's handled in _all_tags.

@fkiraly this is part of your current work around, but introduces less unnecessary intermediate code and instead uses the tags as we would use them later, no?

So I suggest to merge this and then go ahead with the refactoring of the concrete forecasters.

hm, but then, you would have to change the "tags" inspection in _set_fh to _all_tags, no?

that's what's happening inside _has_tag

not sure I understand what you mean, but I mean that current line 524, optfh = self._tags["fh_in_fit"] == "optional" would have to use _all_tags instead, no?

oh, I see, that's already happening inside _has_tag, got it.

I changed that line in this PR, take a look at the suggested changes in this PR. They would be merged into your branch in #912. Does that help?

yes, I just overlooked the change in line 524, all fine