fides - Fides 0.7.8

• Replaced setup.py with pyproject.toml
• added pre-commit hooks for isort & black and applied formatting to all files

- Python
Published by FFroehlich about 3 years ago

fides - Fides 0.7.7

• drop support for python 3.7 and 3.8
• decrease logging verbosity for successful runs

- Python
Published by FFroehlich about 3 years ago

fides - Fides 0.7.6

• adds license and other metadata to setup.py, thanks to @drhagen

- Python
Published by FFroehlich about 3 years ago

fides - Fides 0.7.5

• Adds a notebook containing a short tutorial to use fides for general purpose optimization.
• Slims logging output such that it nicely fits into a single line in jupyter notebooks
• Refactors a trust region radius updating
• Fixes a bug in the convergence check for absolute/relative function tolerance

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.7.4

• Fixes affine scaling for bounded problem. This fixes a critical bug introduced in 0.7.2, updating is strongly recommended

Full Changelog: https://github.com/fides-dev/fides/compare/0.7.3...0.7.4

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.7.3

• reenables specification of stepback and subspace strategies via appropriate strings

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.7.2

• adds additional option value type validation
• fixes runtime warning for unbounded problems
• fixes optimization based stepback mode refined

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.7.1

• Fixes a bug in tracking that incorrectly indicated non-updating in hessian approximation
• Fixes a bug where curvature condition in BFGS approximation was not ignored even when corresponding option was enabled
• Fix subspace normalization for cases when newton step was within trust region radius
• Extend tracing by condition numbers

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.7.0

• Add option to track history of optimization for detailed debugging
• Add options to customize BFGS (and derived) approximation schemes by ignoring curvature condition
• Add novel hybrid method that switches approximations based on fraction of steps with inconclusive trust region ratio
• Remove option to include scaled gradient as candidate step
• Refactor step refinement and restrict it to subspace of base step
• Reduce logging output verbosity

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.6.3

• simplify and fix structured secant method based hybrid schemes
• removes the PSB approximation scheme

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.6.2

• Add citation.cff file
• update description
• bugfixes for hybrid hessian approximation schemes

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.6.1

• fix and better checks for Hessians with NaN values
• add citation information

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.6.0

• Adds new rank 1/2 approximations (BB, BG, PBS, Broyden)
• Adds hybrid hessian approximation schemes (FX, GNSBFGS, SSM, TSSM, HybridUpdate renamed to HybridFixed)
• Adds the possibility to specify residual functions for non-linear least-squares problems, permitting better exploitation of problem properties.

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.5.1

• Changes the default subspace method to 2D by default

- Python
Published by FFroehlich over 4 years ago

fides - Fides 0.5.0

• Disable scaled gradient by default
• Change hybrid implementation from predefined iteration to dynamic scheme

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.4.2

• Fix gradient relative tolerance check for negative objective function values
• Updates to README
• Allow disabling scaled gradient steps

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.4.1

• Fix a bug where lanczos method would yield complex optimization variables
• Update readme

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.4.0

• use lanczos method to compute better descent solution if newton is not a descent direction (2D subspace minimization only)
• add additional safeguards for step acceptance and tr update
• add additional scaled gradient to stepback strategy
• add steinhaugs cg method as subspace solver
• fix 1d trust region analytical solution
• update documentation (thanks to @dweindl)

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.3.4

• Allow initialization of BFGS/SR1 with user provided hessian
• Restrict refinement step-back strategy to only be applied to the best proposal step

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.3.3

• Fixes an issue where the truncated base step was erroneously not included in the set of stepback proposals

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.3.2

• Change x convergence criteria

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.3.1

• Fix convergence checks and stop optimization when tr radius becomes too small
• Change default switch iteration for Hybrid Hessian Approximaiton
• Check for degeneracy in SR1 updates

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.3.0

• Adds new Hybrid Updating Scheme that generates an approximation from the start, but only applies is after a certain number of iterations.

- Python
Published by FFroehlich about 5 years ago

fides - Fides 0.2.5

• Changes default tolerances for FATOL and FRTOL
• Fixes 2D subspace computation, this previously led to inaccurate results for parameters close to the boundary.
• Fixes the trust region radius updating scheme, previously internal solutions to the trust region subproblem would force radius decrease.

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.2.4

• Changes acceptance criteria to accept steps independently of trust region refinement

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.2.3

• Improve 2D subspace performance for ill-conditioned problems

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.2.2

• Refactored code organization
• Improved Hessian Update initialization
• Refactored and improved logging
• Fixed bugs for empty subspaces and singular hessians

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.2.1

• Updates default algorithms after evaluation on benchmarks
• Adds single_reflect strategy
• Some minor updates to logging

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.2.0

• Fixes another bug for 0-dimensional subspaces
• Makes mixed truncation/reflection default option
• Adds option to refine steps via trust-constr optimization (enabled by default)
• Removes convergence based on too small delta
• Adds several dimension checks that should better safeguard against malformed user input.

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.1.1b

• Fixed a bug to handle empty subspaces more graciously
• No longer require setting of dimensionality when initializing HessianApproximation
• Add a couple more dimensionality checks

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.1.1a

• Fixed a bug to handle empty subspaces more graciously
• No longer require setting of dimensionality when initializing HessianApproximation
• Add a couple more dimensionality checks

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.1.0

• Better documentation of options
• Made reflective linesearch recursive and added truncated step-back strategy
• Make it possible to modify previously hardcoded parameters through options
• Add exitflags
• Proper tracking of minima
• Bugfixes for stepback and 1D linesearch

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.0.2

Improves documentation and fixes a critical bug that led to wrong 1d subspace subspace solutions, leading to incorrect gradient and reflected steps.

- Python
Published by FFroehlich over 5 years ago

fides - Fides 0.0.1

Initial Release

- Python
Published by FFroehlich over 5 years ago