FrankWolfe.jl

FrankWolfe v0.3.7

Diff since v0.3.6

Merged pull requests:

• breaking change on step type (#483) (@matbesancon)

FrankWolfe v0.3.6

Diff since v0.3.5

Merged pull requests:

• Bug fix lazy BPCG step (#482) (@JannisHal)

FrankWolfe v0.3.5

Diff since v0.3.4

Merged pull requests:

• added doc for pfw to docs (#459) (@matbesancon)
• Remove abs smooth file. (#462) (@dhendryc)
• Extend ALM and BCFW documentation and add Multi-Threading (#463) (@JannisHal)
• Nonnegative K-sparse polytope / hypersimplex (#464) (@matbesancon)
• Quadratic specialized active set (#466) (@sebastiendesignolle)
• ConvexHull is a subtype of LMO (#470) (@matbesancon)
• Ordered Weighted Norm l1 LMO (#471) (@victorthouvenot)
• Fix ALM reporting issue (#472) (@JannisHal)
• Add limit to UpdateOrders (#473) (@JannisHal)
• Warm-start quadratic active set (#474) (@sebastiendesignolle)
• Bell example literate (#475) (@sebastiendesignolle)
• Alm lambda update (#476) (@JannisHal)
• Add symmetric arrays (#477) (@sebastiendesignolle)
• Symmetric birkhoff example (#478) (@sebastiendesignolle)
• Lazy BPCG step (#479) (@JannisHal)
• Weight purge threshold (#480) (@sebastiendesignolle)
• move gradient function (#481) (@matbesancon)

Closed issues:

• Python interface as in odesolve.jl (#30)
• Non-Euclidean / non-vector space examples (#66)
• function and gradient interface (#133)
• basic benchmarking (#141)
• Issue with bigfloat in away-step frank-wolfe (#203)
• Frank-Wolfe patch release might have broken InferOpt tests (#387)
• Ordered Weighted Norm (#409)
• Quadratic specialized ActiveSet (#461)
• Reporting issue with ALM algorithm (#468)
• ConvexHullOracle not a subtype of LinearMinimizationOracle (#469)

FrankWolfe v0.3.4

Diff since v0.3.3

Merged pull requests:

• eps in Adaptive now working for complex iterates (#455) (@sebastiendesignolle)
• Add symmetry reduction (#456) (@sebastiendesignolle)
• Implementation of pairwise FW (#457) (@elwirth)
• fix redundant constructor in symmetry (#458) (@matbesancon)

FrankWolfe v0.3.3

Diff since v0.3.2

Merged pull requests:

• Implement additional plot for infeasibility (ALM) (#452) (@JannisHal)
• Add callback state and dual gaps to UpdateOrder paramters (#453) (@JannisHal)
• creating branch modern_olss (#454) (@elwirth)

FrankWolfe v0.3.2

Diff since v0.3.1

Merged pull requests:

• ALM and BCFW update (#450) (@JannisHal)

FrankWolfe v0.3.1

Diff since v0.3.0

Merged pull requests:

• abstract active set (#445) (@matbesancon)
• change return type to NamedTuple (#447) (@matbesancon)
• drop step AFW (#449) (@matbesancon)

Closed issues:

• remove using XYZ when not used in examples (#446)

FrankWolfe v0.3.0

Diff since v0.2.38

Merged pull requests:

• fix call (#439) (@matbesancon)
• Fix call (#440) (@matbesancon)
• agnostic printing (#441) (@matbesancon)
• added iterate computation for sparse matrix (#442) (@matbesancon)
• Update list of methods for custom iterate (#443) (@sebastiendesignolle)
• New line searches (#444) (@pokutta)

Closed issues:

• Relax assumption that gradient is in-place (#238)
• Include shape-preserving block vector (#412)
• Github PR template (#424)
• change reporting of agnostic step size to include the "\ell" in bracket (#432)
• Release (#438)

FrankWolfe v0.2.38

Diff since v0.2.37

Merged pull requests:

• 429 almo and bcfw changes (#437) (@JannisHal)

Closed issues:

• ALMO and BCFW changes (#429)

FrankWolfe v0.2.37

Diff since v0.2.36

Merged pull requests:

• block oracles (#434) (@matbesancon)

FrankWolfe v0.2.36

Diff since v0.2.35

Merged pull requests:

• more control over plots (#423) (@matbesancon)
• shorter batch name in header (#433) (@matbesancon)
• Fix Shortstep for zero direction (#435) (@JannisHal)
• convex hull LMO (#436) (@matbesancon)

FrankWolfe v0.2.35

Diff since v0.2.34

Merged pull requests:

• Mostly cosmetic changes and an example (#431) (@pokutta)

Closed issues:

• Homogenize algorithm headers (#430)

FrankWolfe v0.2.34

Diff since v0.2.33

Merged pull requests:

• added open loop parameterization (#428) (@matbesancon)

Closed issues:

• Open-loop step size (#427)

FrankWolfe v0.2.33

Diff since v0.2.32

Merged pull requests:

• better error (#422) (@matbesancon)

Closed issues:

• Optimize all step size computation strategies (#42)
• In memory mode we write back into the x0 (#59)
• Using reference tests for extended verifications (#95)
• Rewrite LCG as a call to FrankWolfe with the suitable LMO (#213)
• Unclear error in active_yet_argminmax (#421)
• Interface and examples for infinite-dimensional optimization (#426)

FrankWolfe v0.2.32

Diff since v0.2.31

Merged pull requests:

• smoothness fallback (#417) (@matbesancon)
• Alm minor fixes (#418) (@JannisHal)
• Plot trajectories marker (#419) (@JannisHal)
• Primal offset and documentation update (#420) (@JannisHal)

FrankWolfe v0.2.31

Diff since v0.2.30

Merged pull requests:

• Implement Alternating Linear Minimization (#407) (@JannisHal)
• Update basics.md (#410) (@pokutta)
• Fix docstrings for Alternating Linear Minimization (#411) (@JannisHal)
• dense gradient by default (#413) (@matbesancon)
• Alternative relaxed smoothness condtion for adaptive linesearch (#415) (@JannisHal)
• Adaptive linesearch update docstring (#416) (@JannisHal)

FrankWolfe v0.2.30

Diff since v0.2.29

Merged pull requests:

• Fixes step selection issue in lazy blended pairwise CG (#408) (@pokutta)

FrankWolfe v0.2.29

Diff since v0.2.28

Merged pull requests:

• reshape birkhoff vector (#390) (@matbesancon)
• inplace version in away fw step (#405) (@matbesancon)

FrankWolfe v0.2.28

Diff since v0.2.27

Merged pull requests:

• Afw inplace (#402) (@matbesancon)
• Copyto scaledhot (#403) (@matbesancon)

FrankWolfe v0.2.27

Diff since v0.2.26

Merged pull requests:

• modified fallback (#401) (@matbesancon)