Christopher Dupuis

1\) IME, batch generation does seem slow, I saw a major slowdown when I implemented it in my current project. 4\) It's probably just taking a long time, I'm seeing...

Sounds good, I'll try adding the squeeze option and see what happens. I'm guessing xbatcher is supporting non-Keras use cases as well?

Still need to add unit test(s)

@jhamman @rabernat I think this is ready to be merged

Doing some cleanup atm

@jhamman @rabernat Sorry for the confusion, I think this is in a mergeable state now.

I think you could actually run the code [here](https://github.com/pangeo-data/xbatcher/issues/38#issuecomment-970673239) with vanilla xbatcher to see the problem.

I understand that, but I don't think the example I posted can be reduced much more than it is. I've stripped out as much as I can and updated it,...

With the changes to the example in #38, I think the reproducing example here has diverged slightly. This should demonstrate the dimension mismatch with the current xbatcher code. I think...

It appears that this issue only arises when _all_ dimensions are input dimensions. Otherwise, there will be a non-trivial batch dimension that works as initially expected.