v0.2.0

New Features

• Allow for running each iteration in a bootstrap multiple times with different fixed pairs (#9)

  • Implemented with a new n_repeats kwarg for DiffPaSSModel.fit_bootstrap
  • By performing several repeats of each bootstrap iteration, we can greedily select the best repeat by hard loss, and use that repeat to select the next set of fixed pairs. This should improve performance in hard cases.

• New tutorial notebook on graph alignment, covering diffpass.train.GraphAlignment and using n_repeats in fit_bootstrap (#11)

v0.1.1

Breaking Changes

• Store hard and soft losses as Python scalars instead of 0-dimensional NumPy arrays (#3)

New Features

• Unify type annotations for group_sizes (#7)

• Add possibility to include diagonals in IntraGroupSimilarityLoss computations (#5)

• Store hard and soft losses as Python scalars instead of 0-dimensional NumPy arrays (#3)

Bugs Squashed

• Fix fit_bootstrap appending empty lists (#1)

v0.1.0

Breaking Changes

• Change signature of get_robust_pairs (ebb160c)

New Features

• Add remove_groups_not_in_both function (876d017)

Full Changelog: 0.0.1...0.1.0

v0.0.1

First DiffPaSS public release.