Sasha Rush

Gotcha. I wish it just hid `_property` automatically, maybe I'll switch to __ for constants.

Any updates?

Yes, we've been debating this point. The current aggressive proposal is to change shape to not be an ordereddict at all, and remove the transpose function or have it return...

Looks great, but lint tests are failing. Be sure to run format.sh first.

Also be careful. MaxUnpool needs to indices, so it is not going to work. ``` input: the input Tensor to invert indices: the indices given out by MaxPool3d ```

Changed around the nn style a bit. maybe we can update avg pool to the new style. (honestly though we won't use any of these this assignment, they're more for...

Note to self: ntorch is a metaclass and intercepts method calls. To do this, there must be some way to make a meta-module.

What a fun question, I hadn't thought of this at all. What if we made a helper wrapper that caught RuntimeError's and "find/replaced" => "dimension \d" with the name?

For cases like MM hopefully we can handle the error before it hits Torch. Indexing cases are a bit harder, although maybe we can check range values in the wrapper.

Mostly done with https://github.com/harvardnlp/namedtensor/pull/88