Cotengra Versions Save

Hyper optimized contraction trees for large tensor networks and einsums

8 months ago

What's Changed

faster index computations, and candidate faster greedy and optimal implementations
allow single term pre-processing in order to support arbitrary einsums
change 'flops' to everywhere be scalar operations not specialised to real float dtypes (results in halving cost in various places)
more preparation to fully decouple from opt_einsum
more robust path caching for many parallel processes
add utils.perverse_equation and utils.tree_equation
add CI testing
Adding fallback for kahypar version number check by @emprice in https://github.com/jcmgray/cotengra/pull/25

Full Changelog: https://github.com/jcmgray/cotengra/compare/v0.3.0...v0.3.1

9 months ago

add ContractionTree.slice_key
default to cotengra bmm implementation for numpy only
unify scoring objectives in scoring.py
update compressed contraction path finding
centralize and internalize necessary opt_einsum functionality to make it an optional dep
add ContractionTree.get_eq_sliced and friends
remove obsolete SlicedContractor
add hypergraph.py for hypergraph functionality
use quimb not hypernetx for rubberband plots
fix compressed contraction missing index bug
fix contract_expression for single terms
allow pairwise contractions with 52+
initial support for cuquantum contraction
add lazy output chunked example
add approx contraction example using quimb
make ContractTreeCompressed metrics default to compressed versions
suppress kahypar warning by removing dangling indices pre partitioning

Full Changelog: https://github.com/jcmgray/cotengra/compare/v0.2.0...v0.3.0

1 year ago

Initial relase for pypi.

Fix link to tqdm by @RSBat in https://github.com/jcmgray/cotengra/pull/6
Allow colormap instances to plot functions by @emprice in https://github.com/jcmgray/cotengra/pull/7
add how to install from source by @Z-Y00 in https://github.com/jcmgray/cotengra/pull/9

Full Changelog: https://github.com/jcmgray/cotengra/commits/v0.2.0