The problems we typically solve in our group are nonconvex, nonlinear, constrained, gradient-based, often computationally expensive, and sometimes have sparse Jacobians. This package wraps derivative computation methods and optimization solvers that are well-suited to these types of problems.
Features:
- Allows easy switching between SNOPT and IPOPT from a common interface passing through all solver options, preserving output in files, and allowing warm starts (for SNOPT).
- Easy switching between various differentiation methods: ForwardDiff, ReverseDiff, Zygote, FiniteDiff (forward, central, complex step), and user-defined derivatives.
- Derivative calculations are all non-allocating during optimization.
- Outputs are also cached as applicable to avoid unnecessary function calls.
- Methods are provided to help determine sparsity patterns, sparse Jacobians can be updated efficiently with SparseDiffTools (using graph coloring), and the sparsity structure is passed to the solvers.
To Install
] add SNOW