Make index check on statespace data less strict

[jessegrabowski]

Closes #384

I was assuming that any pandas index of integers was a RangeIndex; this will now accept anything.

One issue is that if the index has holes (like the index goes 0, 1, 3, 4, ...), it will just be totally ignored, and internally we'll treat it as though it was continuous. I might be able to check for this, but it might also be the type of thing that you have to trust users to do the right thing with. Open to suggestions.

[ricardoV94]

For the holes can you check if len = (max - min + 1)?

[AlexAndorra]

Thanks @jessegrabowski ! Looks good, modulo Ricardo's comment.

One issue is that if the index has holes (like the index goes 0, 1, 3, 4, ...), it will just be totally ignored, and internally we'll treat it as though it was continuous. I might be able to check for this, but it might also be the type of thing that you have to trust users to do the right thing with. Open to suggestions.

Should we error out and tell users they should pass in the missing data points so that they are inferred? Or at least display a warning that highlights that with this behavior, all data points are assumed to be linearly spaced?

[jessegrabowski]

The thing already warns pretty aggressively, so I'm not really inclined to go that route (if anything I think we should go the other way and remove some of the useless warnings).

I don't have a good sense of whether an index [1, 2, 3, 5] is a valid time series index. What to do hinges on the answer to that question.

[AlexAndorra]

I don't have a good sense of whether an index [1, 2, 3, 5] is a valid time series index

My intuition is that it's a shorthand for [1, 2, 3, NaN, 5], so if statespace treats it like [1, 2, 3, 4] , we might want to error out and ask the user to provide the explicit time index, i.e [1, 2, 3, 4, 5]. Does that make sense?