v0.6.0

InfiniteOpt v0.6.0

Diff since v0.5.9

Breaking changes

• NLPExpr is dropped in favor of JuMP.GenericNonlinearExpr
• @register has been replaced with JuMP.@operator
• map_nlp_to_ast has been discontinued
• Optimizer model API has been replaced with the more general transformation backend API. For modelling, the syntax is largely the same, but accessing methods like optimizer_model_variable have been changed to transformation_variable. The old API is still supported via deprecation and users are encouraged to run Julia in deprecation mode to update their code.
• DomainRestrictions have been replaced with DomainRestriction which enforces restrictions based on arbitrary Julia functions (similar to parameter functions). Please see the constraint guide in the documentation for details.
• start_value_function and set_start_value_function have been dropped in favor of JuMP.start_value and JuMP.set_start_value.
• Semi-infinite variables have been refactored internally to behave more consistently with point variables. The user API remains the same, though bounds and start values can now be specified.
• Point variables inherit domain info from infinite variables in a more general way. Semi-infinite and point variables defined via macros define their own info that overwrites that of the infinite variable. Semi-infinite and point variables defined functionally only modify info if specified via domain modification methods (e.g., set_lower_bound, delete_lower_bound).
• @infinite_parameter no longer supports DenseAxisArrays and SparseAxisArrays in favor of Array to significantly improve performance.
• Collections.VectorTuple have been refactored to focus on tuples of Arrays and Numbers.
• Core data structures have been modified to simply the code base; however, users are not encouraged work directly with core data objects.

Features

• Trained machine learning models can be embedded via the InfiniteMathOptAI extension.
• Transcribed values can be converted to interpolated functions via the InfiniteInterpolations extension.
• Semi-infinite variables can now have bounds and start values.
• Higher-order derivatives are now preserved to facilitate more accurate transformations.
• Recursion is removed to handle deeply nested nonlinear expressions without stackoverflow errors
• Non-JuMP models (e.g., ExaModels) can now be interfaced via the new transformation API.
• set_parameter_value now preserves the backend for efficient resolves.
• warmstart_backend_start_values efficiently warmstarts the backend for resolves.
• Support is added for MOI.Parameter.
• New examples have been added to the documentation.
• Misc. bug fixes and documentation improvements.

Merged pull requests:

• Refactor to directly support higher order derivatives (#341) (@pulsipher)
• Fix SigFig Bug (#344) (@pulsipher)
• Improve the Extensibility of Derivative Methods (#345) (@pulsipher)
• Support Expression Restrictions (#348) (@pulsipher)
• Generalize Transformation API (#349) (@pulsipher)
• Backend API Refinements (#351) (@pulsipher)
• Remove ndarray Keyword Argument (#356) (@pulsipher)
• Use OrderedDict for Array Parameter Supports (#357) (@pulsipher)
• Allow Non-Vector Input for Infinite Parameter Supports (#358) (@pulsipher)
• Improve Performance of Internal Semi-Infinite Variables (#359) (@pulsipher)
• Fix bug in restricted_copy (#360) (@pulsipher)
• Fix printing test for JuMP v1.23.0 (#361) (@odow)
• Improve Defaults of JuMP.is_binary, JuMP.is_integer, and JuMP.is_fixed (#365) (@pulsipher)
• Use invokelatest after eval to adjust to upcoming julia change (#369) (@Keno)
• Addition of three new examples under optimial control. (#370) (@dnguyen227)
• Fix Deletion of Variable Dependencies (#377) (@pulsipher)
• Update CI Versioning (#378) (@pulsipher)
• Update Optimal Control Example Documentation (#380) (@wenwen0231)
• Add InfiniteInterpolate as an extension (#382) (@wenwen0231)
• CompatHelper: bump compat for DataStructures to 0.19, (keep existing compat) (#385) (@github-actions[bot])
• Map finite parameters to JuMP.Parameters (#387) (@wenwen0231)
• Map parameter functions to JuMP.Parameters (#389) (@wenwen0231)
• Add Extension for MathOptAI (#390) (@pulsipher)
• Properly Handle Parameter Functions as JuMP.Parameters in Measures (#391) (@wenwen0231)
• Updated Versioning (#392) (@pulsipher)
• Clean Up and Simplify Dependence on Infinite Parameters (#393) (@pulsipher)
• Improve Parameter Support (#394) (@pulsipher)
• Add Support for Functions in Variable Domains and Overhaul Domain Restrictions (#395) (@pulsipher)
• Provide an API for Reformulating High Order Derivatives into 1st Order Derivatives (#396) (@pulsipher)
• Remove Parameter Numbers (#397) (@pulsipher)
• Remove Recursion from map_expression (#398) (@pulsipher)
• Prep for v0.6 (#399) (@pulsipher)

Closed issues:

• Cannot Transcribe Scalar Measures/Derivatives of Dependent Infinite Parameters (#56)
• Return Interpolated Value Functions (#82)
• Point Variable Inheritance is Invalidated with Changes to Infinite Variable Info (#148)
• Specify Exclusion Points for Constraints (#167)
• Maintain Higher-Order Derivative Representations (#197)
• Vector-Valued Nonlinear Functions (#229)
• Restrict Infinite Parameters to only use Arrays (#233)
• Allow Functional Bounds on Infinite Variables (#253)
• Support Restrictions on Expressions (#347)
• Can i write the InfiniteOpt's Model to file of nl? (#364)
• [FEATURE] Query expression and parameter function names. (#367)
• Deleting variables does not work when a constraint evaluates a variable at multiple time points (#376)
• Update example documentation (#379)
• Ability for Callback function during optimization run (#383)
• Can't install/precompile InfiniteOpt (#388)