- Changed the default tolerance for the Frank–Wolfe gap to 1e-16
- Removed row-stochasticity check after each iteration for extra speed
- Fixed issue with last iterations in a restart not counting towards iteration budget
- Fixed issue with zero iteration budgets when using c='auto'
- Improved readability of the example
- Renamed
ub()
toupper()
- Corrected random number generator state in the search for best
$c$ and its effects on the subsequent computations
- Added link to the paper
- Added "Basics" section
- Layout and readability improvements
- Added searching for best
$c$ as default behavior - Added function for random generation of a bi-mapping polytope vertex
$\mathbf{R}$ - Added
__version__
variable - Python >= 3.8 is now required
- Removed convexity-based calculation of
$c$ and warm-start sequences of$c$ due to inefficiency
- Created "Advanced" section where the logic behind parameter
$c$ is explained
- Added
$(f, g)$ to verbose output if returning the mappings
- Corrected (halved) ub in summary (verbose > 0)
- Corrected project homepage
- Switched to url for the example illustration
- Simplified import structure
- Added "summary" verbose level
- Updated example illustration
- Enabled setting lower bound to avoid redundant iterations
- Added flag for validating the triangle inequality
- Fixed import structure
- Distances are scaled to prevent overflow
- Corrected library import in README.md
- Removed redundant
__init__.py
- Renamed source directory
- Allowed for importing (oops)
- Corrected dimensions in
fg_to_R
- Different levels of verbosity
- Expanded README.md
- Prevented overflow from c**2
- Switched to global iteration budget with no limitations per restart
- Added this changelog
- First release of
dgh