Add needs_bounds and supports_infinite_bounds fields in algo info

[gauravmanmode]

PR Description

This PR references issue #597.
Currently, there is no easy way to find out if an algorithm requires bounds. Not all algorithms that support supports bounds also requires them.

Proposed Solution:

• Add a needs_bounds and supports_infinite_bounds field analogus to suuports_bounds in the mark.minimizer decorator.

• Optional: Strict checking of needs_bounds and supports_infinite_bounds fields in optimize.py.

• This requires a new exception for error handling, namely:

  • IncompleteBoundsError if bounds are required but not provided or if infinite value is encountered in parameter bounds.

This table lists values for each optimizer that supports bounds.

Show Table

type	name	needs_bounds	supports_infinite_bounds	reason
LOCAL	fides	False	True	algo works on unconstrained problems, set to -np.inf and np.inf for unbounded
LOCAL	iminuit_migrad	False	True	"setting bounds makes a transfom of internal parameters like optimagic, set to None if inf values"
LOCAL	ipopt	False	True	it will set to -np.inf and np.inf itself
LOCAL	nag_dfols	False	True	algo works with unconstrained var; bounds not compulsory
LOCAL	nag_pybobyqa	False	True	"
GLOBAL	nevergrad_bo	False	False	samples from N(0,1)
GLOBAL	nevergrad_cga	False	False	"
GLOBAL	nevergrad_cmaes	False	False	"
GLOBAL	nevergrad_de	False	False	"
GLOBAL	nevergrad_eda	False	False	"
GLOBAL	nevergrad_emna	False	False	"
GLOBAL	nevergrad_meta	False	False	"
GLOBAL	nevergrad_NGOpt	False	False	"
GLOBAL	nevergrad_oneplusone	False	False	"
GLOBAL	nevergrad_pso	False	False	"
GLOBAL	nevergrad_randomsearch	False	False	"
GLOBAL	nevergrad_samplingsearch	False	False	"
GLOBAL	nevergrad_tbpsa	False	False	"
LOCAL	nlopt_bobyqa	False	True	If a lower/upper bound is not set, the default is no bound (unconstrained, i.e. a bound of infinity)
LOCAL	nlopt_ccsaq	False	True	"
LOCAL	nlopt_cobyla	False	True	"
GLOBAL	nlopt_crs2_lm	True	False	Only bound-constrained problems are supported by this algorithm.
GLOBAL	nlopt_direct	True	False	"
GLOBAL	nlopt_esch	True	False	"
GLOBAL	nlopt_isres	True	False	"
LOCAL	nlopt_lbfgsb	False	True	If a lower/upper bound is not set, the default is no bound (unconstrained, i.e. a bound of infinity)
LOCAL	nlopt_mma	False	True	"
LOCAL	nlopt_newuoa	False	True	"
LOCAL	nlopt_neldermead	False	True	"
LOCAL	nlopt_slsqp	False	True	"
LOCAL	nlopt_sbplx	False	True	"
LOCAL	nlopt_tnewton	False	True	"
LOCAL	nlopt_var	False	True	"
LOCAL	pounders	False	True	uses normalized bounds. If none provided, uses unit cube [-1, 1]
GLOBAL	pygmo_bee_colony	True	False	need finite bounds for sampling
GLOBAL	pygmo_cmaes	True	False	"
GLOBAL	pygmo_compass_search	True	False	"
GLOBAL	pygmo_de	True	False	"
GLOBAL	pygmo_de1220	True	False	"
GLOBAL	pygmo_gaco	True	False	"
GLOBAL	pygmo_gwo	True	False	"
GLOBAL	pygmo_ihs	True	False	"
GLOBAL	pygmo_mbh	True	False	"
GLOBAL	pygmo_pso	True	False	"
GLOBAL	pygmo_pso_gen	True	False	"
GLOBAL	pygmo_sade	True	False	"
GLOBAL	pygmo_sea	True	False	"
GLOBAL	pygmo_sga	True	False	"
GLOBAL	pygmo_simulated_annealing	True	False	"
GLOBAL	pygmo_xnes	True	False	"
GLOBAL	scipy_basinhopping	False	True	local optimizers used by algo work on unconstrained/bounded vars
GLOBAL	scipy_brute	True	False	need finite bounds
GLOBAL	scipy_differential_evolution	True	False	"
GLOBAL	scipy_direct	True	False	"
GLOBAL	scipy_dual_annealing	True	False	"
LOCAL	scipy_lbfgsb	False	True	algo works with unconstrained prob
LOCAL	scipy_ls_dogbox	False	True	"
LOCAL	scipy_ls_trf	False	True	"
LOCAL	scipy_neldermead	False	True	"
LOCAL	scipy_powell	False	True	"
GLOBAL	scipy_shgo	False	True	bounds is not a requirement for shgo, mentioned in impl paper
LOCAL	scipy_slsqp	False	True	"
LOCAL	scipy_truncated_newton	False	True	"
LOCAL	scipy_trust_constr	False	True	"
LOCAL	tao_pounders	False	True	set to -np.inf , np.inf
LOCAL	tranquilo	False	True	set to -np.inf , np.inf
LOCAL	tranquilo_ls	False	True	"

For optimizers that do not support bounds, these additional fields will be set to False.

Todo

• Add corresponding fields in the mark.minimizer decorator for every algorithm.
• Explain usage in the How-to-Guide on how to choose algorithms.
• Add additional tests if required

Closes #597

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov]

Codecov Report

Attention: Patch coverage is 86.66667% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/optimagic/optimization/optimize.py	75.00%	2 Missing ⚠️

Files with missing lines	Coverage Δ
src/optimagic/exceptions.py	88.23% <100.00%> (+0.35%)	⬆️
src/optimagic/mark.py	90.00% <ø> (ø)
src/optimagic/optimization/algorithm.py	94.94% <100.00%> (+0.15%)	⬆️
src/optimagic/optimizers/bhhh.py	90.38% <ø> (ø)
src/optimagic/optimizers/fides.py	95.77% <ø> (ø)
src/optimagic/optimizers/iminuit_migrad.py	97.29% <ø> (ø)
src/optimagic/optimizers/ipopt.py	96.92% <ø> (ø)
src/optimagic/optimizers/nag_optimizers.py	93.30% <ø> (ø)
src/optimagic/optimizers/neldermead.py	92.24% <ø> (ø)
src/optimagic/optimizers/nevergrad_optimizers.py	97.67% <ø> (ø)
... and 7 more

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[janosg]

About iminuit_migrad you wrote:

setting bounds makes a transfom of internal parameters like optimagic, set to None if inf values"

For optimagic users we only want inf to express missing bounds for some parameters, not None. Were you only referring to the original optimizer or is this about the optimagic wrapper?

[janosg]

Thanks for the PR. Looks really good. I have a few questions we can discuss during our next call.

[gauravmanmode]

Hi @janosg , I have made the suggested changes. Can you tell me if I'm missing anything?

[janosg]

This is ready to go. Thanks a lot for this great PR!