2.5.0

• Many speed-ups to SimulationProblem. Partially because of storing things in lists instead of np.arrays (that are expensive to append to).
• Support for string parameters in Modelica models (parameter string foo = "bar").
• HomotopyMixin: Add support for specifying starting theta (e.g. not starting at 0.0, but starting at 0.5 or 1.0)
• Timeseries: Add __eq__ for easy equality comparison of Timeseries (timeseries_a == timeseries_b)

2.4.0

Semi-bugs/features:

• Also allow nominals for path and extra variables
• Many fixes related to (optional) block interpolation of variables

Improved checks:

• Simulation: Check rootfinder convergence using solver return status

New features:

• New mixin to minimize the absolute value of a function or variable (MinAbsGoalProgrammingMixin)
• New mixin to approximate high order penalties in a linear fashion (LinearizedGoalProgrammingMixin)
• New mixin to read and write from NetCDF files (NetCDFMixin)
• Convenience method for merging bounds (OptimizationProblem.merge_bounds)
• Allow passing of arguments to problem class via run_optimization/simulation_problem
• Allow passing of model/input/output folder paths to run_optimization/simulation_problem

Deprecations:

• Deprecate explicit collocation
• Deprecate integrated states

2.3.0

Semi-bugs/features:

• scale_by_problem_size for path goals now only uses number of active time steps.
• Much improved scaling of initial derivatives.
• Improve branch allocation of ControlTreeMixin.
• Fix inconsistencies in internal API w.r.t. what methods can return symbolic values.

Improved checks:

• Sanity checks on goals, e.g. function ranges. Some models may fail the checks now, but if they do the goal in question did not make much sense anyway.

New features:

• Vector goals (mostly a performance improvement)
• Optional more generic/optimal way of translating goals from current priority to the next. See keep_soft_constraints option added in commit f6eb00d.
• Can now provide a custom (initial) seed to SimulationProblem.
• Allow extra variables to appear in path expressions
• New examples, in particular showing the use (and strength) of using HomotopyMixin for channel flow.
• Many large and small optimizations and refactorings for perform

2.2.5

This will be second to last bugfix release for the 2.2 series.

2.3.0b1

This marks the feature freeze of the 2.3 branch. Below is a short overview of what has changes with respect to 2.2:

• Semi-bugs/features:
  • scale_by_problem_size for path goals now only uses number of active time steps.
  • Much improved scaling of initial derivatives.
  • Improve branch allocation of ControlTreeMixin.
  • Fix inconsistencies in internal API w.r.t. what methods can return symbolic values.
• Improved checks:
  • Sanity checks on goals, e.g. function ranges. Some models may fail the checks now, but if they do the goal in question did not make much sense anyway.
• New features:
  • Vector goals (mostly a performance improvement)
  • Optional more generic/optimal way of translating goals from current priority to the next. See keep_soft_constraints option added in commit f6eb00d.
  • Can now provide a custom (initial) seed to SimulationProblem.
  • Allow extra variables to appear in path expressions
• New examples, in particular showing the use (and strength) of using HomotopyMixin for channel flow.
• Many, many small optimizations and refactorings for performance reasons. Especially useful for large models with many variables.