Daniel Smith

The `dp` algorithm must identify that tensor as a disconnected subgraph to solve first. This is an oversight of the algorithm, but is unlikely to create a meaningful time discrepancy....

This goes a bit with automatic looping over GEMM-like products such as `aij,ajk->aik`. `max_memory` would be hard to satisfy I think, I guess you could try to automatically loop largest...

This goes a bit back to FLOP cost is not the only metric, especially when it comes to GPU's. For that, it seems a weighted metric would help (outer product:...

Cool, this looks like it is coming along well! I wonder if it would be worth have deeper Dask integration to automatically batch this out. The interface prototype seems good,...

@jcmgray Seems we have a good use case for this. After sitting on it a bit, how easy do you think it would be to port the slicing component over?

I like 1), but I want to be fairly cautious about it and it feels like something that can be worked through once the slicing infrastructure is complete. Another item...

Yes to `ChunkFinder`, and `chunked_inds={'a': 1, 'b': 2}` was quite useful when I was using Dask. 'edge chunks' simply referring to the common case of non-integer numbers of chunks for...

We have always focused on generalized expressions, but have not focused on specific networks. Are there other networks beyond Apollonian that we should consider adding to the test suite and...

@jcmgray Do you think we can fix this with current algorithms or should we look at adding flowcutter/quickbb for these highly structured graphs?

I added the greedy and optimal versions of the `opt_einsum` code directly to NumPy a number of years ago (oof, blame says 5!). Parsing through the blame, it hasn't been...