Constants
This module provides a central place to define native python enums and constants that are used in multiple other modules
Classes Summary
|
Defines possible exitflag values for the optimizer to indicate why optimization exited. |
|
Defines all the fields that can be specified in Options to |
|
Defines the possible choices of search refinement if proposed step reaches optimization boundary |
|
Defines the possible choices of subspace dimension in which the subproblem will be solved. |
Functions Summary
|
Check if the chosen options are valid |
Classes
- class fides.constants.ExitFlag(value, names=<not given>, *values, module=None, qualname=None, type=None, start=1, boundary=None)[source]
Defines possible exitflag values for the optimizer to indicate why optimization exited. Negative value indicate errors while positive values indicate convergence.
- DELTA_TOO_SMALL = -5
Trust Region Radius too small to proceed
- DID_NOT_RUN = 0
Optimizer did not run
- EXCEEDED_BOUNDARY = -4
Exceeded specified boundaries
- FTOL = 1
Converged according to fval difference
- GTOL = 3
Converged according to gradient norm
- MAXITER = -1
Reached maximum number of allowed iterations
- MAXTIME = -2
Expected to reach maximum allowed time in next iteration
- NOT_FINITE = -3
Encountered non-finite fval/grad/hess
- XTOL = 2
Converged according to x difference
- class fides.constants.Options(value, names=<not given>, *values, module=None, qualname=None, type=None, start=1, boundary=None)[source]
Defines all the fields that can be specified in Options to
Optimizer- DELTA_INIT = 'delta_init'
initial trust region radius
- ETA = 'eta'
trust region increase threshold for trust region ratio
- FATOL = 'fatol'
absolute tolerance for convergence based on fval
- FRTOL = 'frtol'
relative tolerance for convergence based on fval
- GAMMA1 = 'gamma1'
factor by which trust region radius will be decreased
- GAMMA2 = 'gamma2'
factor by which trust region radius will be increased
- GATOL = 'gatol'
absolute tolerance for convergence based on grad
- GRTOL = 'grtol'
relative tolerance for convergence based on grad
- HISTORY_FILE = 'history_file'
when set, statistics for each start will
- MAXITER = 'maxiter'
maximum number of allowed iterations
- MAXTIME = 'maxtime'
maximum amount of walltime in seconds
- MU = 'mu'
acceptance threshold for trust region ratio
- STEPBACK_STRAT = 'stepback_strategy'
method to use for stepback
- SUBSPACE_DIM = 'subspace_solver'
trust region subproblem subspace
- THETA_MAX = 'theta_max'
maximal fraction of step that would hit bounds
- XTOL = 'xtol'
tolerance for convergence based on x
- class fides.constants.StepBackStrategy(value, names=<not given>, *values, module=None, qualname=None, type=None, start=1, boundary=None)[source]
Defines the possible choices of search refinement if proposed step reaches optimization boundary
- MIXED = 'mixed'
mix reflections and truncations
- REFINE = 'refine'
perform optimization to refine step
- REFLECT = 'reflect'
recursive reflections at boundary
- SINGLE_REFLECT = 'reflect_single'
single reflection at boundary
- TRUNCATE = 'truncate'
truncate step at boundary and re-solve
- class fides.constants.SubSpaceDim(value, names=<not given>, *values, module=None, qualname=None, type=None, start=1, boundary=None)[source]
Defines the possible choices of subspace dimension in which the subproblem will be solved.
- FULL = 'full'
Full \(\mathbb{R}^n\)
- STEIHAUG = 'scg'
CG subspace via Steihaug’s method
- TWO = '2D'
Two dimensional Newton/Gradient subspace
Functions