v0.4.0

Hypatia v0.4.0

Diff since v0.3.0

Closed issues:

• return SlowProgress if some residual norms start increasing too much (#327)
• tighten tols for infeasibility detection check (#347)
• improve termination criteria for outer 6x6 iterative refinement (#375)
• fix argument types of vec to mat functions (#392)
• optimize LMI cone barrier evals (#477)
• for WSOS cones, special case P_r for which L_r = 1 (just use log of 1x1 Lambda instead of logdet) (#533)
• try checking hessian nbhd after heuristic nbhd is satisfied for all cones (#534)
• experiment with higher order correctors (again) (#567)
• MOI interface: set_silent not respected (#640)
• get a more efficient TOA for lmi cone (#654)

Merged pull requests:

• add matrix relative entropy cone (#574) (@lkapelevich)
• fix type issues with interpolation functions (#629) (@chriscoey)
• rename predorcent stepper file (#630) (@chriscoey)
• improve predorcent stepper (#631) (@chriscoey)
• improve predorcent stepper and use curve search (#633) (@chriscoey)
• add more entropy-like cones (#634) (@lkapelevich)
• CompatHelper: bump compat for "IterativeSolvers" to "0.9" (#635) (@github-actions[bot])
• cleanup and add barrier oracles numerics checks (#637) (@chriscoey)
• cleanup and improve slow progress checks (#638) (@chriscoey)
• add third order derivatives for relative entropy (#639) (@lkapelevich)
• add and test stepper options and cleanup search (#641) (@chriscoey)
• if not spawning in runbenchmarks, run compile instance (#642) (@chriscoey)
• simplify steppers and alpha search, add optional initial alpha heuristic (#643) (@chriscoey)
• delete a redundant line of code (#646) (@lkapelevich)
• improve predorcent stepper (#647) (@chriscoey)
• correct MOI.Silent (#648) (@lkapelevich)
• add timings for subprocedures (#649) (@chriscoey)
• use lower bound on hessian nbhd to speed up search (#650) (@chriscoey)
• time factorizations and system solver setup (#652) (@chriscoey)
• remove constant multiplier from rootdet barrier (#653) (@lkapelevich)
• rename randompolymat to nearestpolymat (#655) (@lkapelevich)
• refactor wsos matrix oracles for faster grad (#656) (@lkapelevich)
• refactor wsos L1 cone, making grad faster (#657) (@lkapelevich)
• refactor wsos L2 cone, making grad faster (#658) (@lkapelevich)
• add scripts for benchmarking steppers (#659) (@lkapelevich)
• add various example instances to benchmark (#661) (@lkapelevich)
• get a better LMI TOA (#662) (@lkapelevich)
• faster TOA for wsos nonnegative (#663) (@lkapelevich)
• cleanup cone oracle tests, check oracle allocations (#664) (@chriscoey)
• add power benchmark instances and tweaks in stepper run script (#665) (@lkapelevich)
• speed up matrix wsos TOA (#666) (@lkapelevich)
• WIP: fix some slow oracles and failing instances (#667) (@chriscoey)
• make stepper analysis script produce more CSVs (#668) (@lkapelevich)
• make more tweaks to benchmark analysis scripts (#669) (@lkapelevich)
• improve scalar wsos oracles and test complex case (#670) (@lkapelevich)
• improve LMI cone oracles (#671) (@chriscoey)
• improve wsosmat corrector slightly (#672) (@chriscoey)
• reduce time complexity in the TOO of WSOS L2 cone (#673) (@lkapelevich)
• WIP: use expensive hess prods, only use iterative refinement when needed (#674) (@chriscoey)
• record example type in dataframe (#675) (@chriscoey)
• improve wsos oracles speed (#676) (@chriscoey)
• improve time complexity for WSOS L1 TOO (#677) (@lkapelevich)
• use expensive hessian products only when needed (#678) (@chriscoey)
• count number of instances with loosened tolerances (#679) (@lkapelevich)
• rename shift to back (#680) (@lkapelevich)
• remove unneeded preallocated space in matrix wsos (#681) (@lkapelevich)
• refactor hess prod and correction for wsosmat (#682) (@chriscoey)
• WIP: add different implementation types for sparse PSD cone, add new example (#683) (@chriscoey)
• WIP: add new generic matrix monotone epigraph cone (#684) (@chriscoey)
• delete unused import (#685) (@lkapelevich)