0.26.0

What's Changed

• Manage constraints as functions by @relf in #242

See this example of an optimization with constraint functions compared to the same example where the constraints are provided with the objective (the legacy way) and are approximated with surrogates behind the scene. Constraint functions definition is suited when constraint computation is cheap to avoid surrogate approximation.

To summarize, Egor handles three types of constraint:

1. the ones defined by the bounds of the input domain (hypercube),
2. the ones defined with the objective function which are approximated with surrogates
3. the ones defined as separate cheap-to-compute functions

Notes:

• Gradient definition of a constraint function is required when using SLSQP as infill optimizer
• Constraint tolerance is defined as a float list of length (nb of surrogated constraints p (cstr) + nb of constraints function q (fcstr)) following the order of definition: [cstr1, ... cstr_p, fcstr_1, ..., fcstr_q]

Full Changelog: 0.25.2...0.26.0

0.25.2

What's Changed

• Fix python action by @relf in #229
• Fix working directory in python release action by @relf in #230
• Add readme specific to python egobox binding by @relf in #231
• Re-enable BLAS backend tests by @relf in #234
• Add stub file (type annotations) for Python APIs by @unkcpz in #235
• Fix Python example optim g24 by @relf in #238

New Contributors

• @unkcpz made their first contribution in #235

Full Changelog: 0.25.1...0.25.2

0.25.1.post1

What's Changed

• Fix python action by @relf in #229
• Fix working directory in python release action by @relf in #230

Full Changelog: 0.25.1...0.25.1.post1

0.25.1

What's Changed

• Refactor repository: sub-crates under crates directory by @relf in #226
• Adjust publish workflow by @relf in #227
• Prepare release 0.25.1 by @relf in #228

Full Changelog: 0.25.0...0.25.1

0.25.0

What's Changed

• gp: Breaking change - Change GP training Rust API to use one-dimensional array for output data by @relf in #222
• doe: Fix optimized LHS ESE algorithm by @relf in #219
• egobox::Gpx: Panic if training output data is not one-dimensional by @relf in #218
• Refactor moe::predict_smooth by @relf in #221
• Update to numpy crate 0.22.1 to fix win32 compilation by @relf in #216
• Prepare release 0.25.0 by @relf in #223

Full Changelog: 0.24.0...0.25.0

0.24.post1

What's Changed

• Update to numpy crate 0.22.1 to fix win32 compilation by @relf in #216

Full Changelog: 0.24.0...0.24.post1

0.24.0

What's Changed

• moe: Save/Load surrogates in binary format by @relf in #213
• Fix badge link by @relf in #201
• Upgrade to PyO3 0.22 by @relf in #203
• Bump pyo3 from 0.22.4 to 0.22.5 by @dependabot in #204
• Prepare release 0.24 by @relf in #215

Full Changelog: 0.23.0...0.24.0

0.23.0

What's Changed

• gp:
  • Add training infos getters at Python level by @relf in #196
• ego:
  • Add checkpointing and hot start by @relf in #197
  • Do not activate TREGO by default by @relf in #198
  • Improve hot start API by @relf in #199
• Add EGObox logo by @relf in #193
• Prepare release 0.23 by @relf in #200

Full Changelog: 0.22.0...0.23.0

0.22.0

What's Changed

• ego:
  • Save config and optim history by @relf in #190
  • Move the DOE save after the algo the iteration (fixes #182) by @relf in #185
  • Renaming hotstart option and hist fields of optimization result by @relf in #191
• Maintainance by @relf in #183
• Prepare release 0.22 by @relf in #192

Full Changelog: 0.21.1...0.22.0

0.21.1

What's Changed

• gp: Fix variance gradient computation by @relf in #177

Full Changelog: 0.21.0...0.21.1