CHANGELOG.md

Changes

Version 0.24.0 - 12/11/2024

• 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

Version 0.23.0 - 01/10/2024

• Add EGObox logo by @relf in #193
• 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

Version 0.22.0 - 04/09/2024

• 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
• Maintainance by @relf in #183

Version 0.21.1 - 31/07/2024

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

Version 0.21.0 - 09/07/2024

• ego:
  • Implement TREGO algorithm by @relf in #173
  • Fix added point count in TREGO local step by @relf in #174
  • Fix WB2S criteria scaling factor and fmin computation by @relf in #175
• Prepare release 0.21 by @relf in #176

Version 0.20.0 - 25/06/2024

• gp:
  • Save original parameters in trained gp models by @relf in #166
  • Implement cross validation metric for surrogates by @relf in #167
• ego:
  • Make n_optmod option available in Python by @relf in #161
  • Better Egor solver state handling by @relf in #168
  • Refactor ego module by @relf in #169
• Add dependabot cargo ecosystem check by @relf in #163

Version 0.19.0 - 15/05/2024

• ego:
  • Allow to reuse surrogate trainings (reuse previous hyperparameters) from a previous iteration by @relf in #157
  • Fix hot start mechanism due to bad loaded DOE exploitation by @relf in #156
  • Adjust log messages by @relf in #158
  • Log Egor configuration by @relf in #159
• gp:
  • Allow fixed hyperparameters theta for GP and Sparse GP by @relf in #155
• egobox:
  • Upgrade to pyo3 0.21 by @relf in #152
  • Upgrade dependencies by @relf in #153

Version 0.18.1 - 10/04/2024

• Fix GP mixture with kpls option on Griewank test function by @relf in #150

Version 0.18.0 - 08/04/2024

• [Breaking changes] gp, moe, egobox (Python): Rename predict_derivatives() as predict_gradients() by @relf in #148

Version 0.17.0 - 04/04/2024

• [Breaking changes] gp API renaming by @relf in #145
  • predict_values() is renamed predict()
  • predict_variances() is renamed predict_var()
  • predict_variance_derivatives() is renamed predict_var_derivatives() Derivatives predictions (predict_derivatives() and predict_var_derivatives()) are made available in Python.
• Refactor Mixture of Experts by @relf in #146 Factorize code between full GP and sparse GP implementations

Version 0.16.0 - 07/03/2024

• Add Gpx accessors by @relf in #140
• Fix LHS maximin bug by @relf in #141
• doe: Improve classic, centered and maximin LHS performances by @relf in #138
• doe: Improve optimized LHS performances (1.25x speedup) by @relf in #136
• Rework (mostly internal) API to avoid awkward &Option by @relf in #134
• Add Python bindings for all LHS flavours by @relf in #135

Version 0.15.0 - 02/01/2024

• gp: Implement sparse gaussian process methods (cf. SparseGaussianProcess)
• Python binding: SparseGpMix, see doc/tutorial
• GP/SGP API
  • hyperparameter tuning : initial theta guess and bounds can be specified (theta_init, theta_bounds)
  • n_start controls the number of optimization multistart
• In GP/SGP rayon is used to make parallel optimization multistart

Version 0.14.0 - 13/12/2023

• ego: Fix ask-and-tell interface suggest() method in presence of discrete variable to work in discrete not in continuous space A few API breaking changes:
• EgorConfig::xtypes not an option anymore
• EgorSolver::new_with_xtypes() renamed new as new with xlimits is removed, use to_xtypes to convert xlimits
• EgorConfig::no_discrete attribute removed, use EgorConfig::discrete() method
• SurrogateBuilder::new_with_xtypes_rng renamed new_with_xtypes

Version 0.13.0 - 30/11/2023

• ego: API refactoring to enable ask-and-tell interface
  • Configuration of Egor is factorize out in EgorConfig
  • EgorBuilder gets a configure method to tune the configuration
  • EgorService structure represent Egor when used as service
  • Python Egor API changes:
    • function under optimization is now given via minimize(fun, max_iters-...) method
    • new method suggest(xdoe, ydoe) allows to ask for x suggestion and tell current function evaluations
    • new method get_result(xdoe, ydoe) to get the best evaluation (ie minimum) from given ones

Version 0.12.0 - 10/11/2023

• gp uses pure Rust COBYLA by @relf in #110
• ego as pure Rust implementation (nlopt is now optional) by @relf in #112
• egobox Python module: Simplify mixed-integer type declaration by @relf in #115
• Upgrade dependencies by @relf in #114
• Upgrade edition 2021 by @relf in #109
• CI maintainance by @relf in #111
• Bump actions/checkout from 2 to 4 by @dependabot in #107
• Bump actions/setup-python from 2 to 4 by @dependabot in #108

Version 0.11.0 - 20/09/2023

• Automate Python package build and upload on Pypi from Github CI by @relf in #104
• Fix FullFactorial when asked nb iof samples is small wrt x dimension by @relf in #105
• Make mixed-integer sampling methods available in Python by @relf in #106

Version 0.10.0 - 22/06/2023

• gp, moe and egobox Python module:

  • Added Gaussian process sampling (#97)
  • Added string representation (#98)

• egobox Python module:

  • Change recombination enum to respect Python uppercase convention (#98)

• Notebooks and documentation updates (#97, #98, #99)

Version 0.9.0 - 02/06/2023

• ego:

  • Infill criterion is now a trait object in EgorSolver structure (#92)
  • Egor and EgorSolver API: methods taking argument of type Option<T> now take argument of type T (#94)
  • EgorBuilder::min_within_mixed_space() is now EgorBuilder::min_within_mixint_space() (#96)
  • egobox-ego library doc updated (#95)

• egobox Python module: Upgrade to PyO3 0.18 (#91)

Version 0.8.2 - 31/03/2023

• ego:
  • Fix Egor solver best iter computation (#89)

Version 0.8.1 - 28/03/2023

• ego:
  • Make objective and constraints training in parallel (#86)
  • Lock mopta execution to allow concurrent computations (#84)
  • Fix and adjust infill criterion optimmization retries strategy (#87)
• moe:
  • Fix k-fold cross-validation (#85)

Version 0.8.0 - 10/03/2023

• ego:
  • Renaming XType, XSpec for consistency (#82)
  • Export history in optimization result (#81)
  • Use nb iter instead of nb eval, rename q_parallel as q_points (#79)
  • Warn when inf or nan detected during obj scaling computation (#78)
  • Parallelize constraint scales computations (#73)
  • Parallelize multistart optimizations (#76)
  • Handle GMM errors during MOE training (#75)
  • Handle possible errors from GMM clustering (#74)
  • Upgrade argmin 0.8.0 (#72)
  • Add mopta08 test case as example (#71)
  • Fix scaling check for infinity (#70)
  • Use kriging surrogate by default (#69)

Version 0.7.0 - 11/01/2023

• gp:
  • Add analytic derivatives computations (#54, #55, #56, #58, #60). All derivatives available for all mean/correlation models are implemented.
  • Refactor MeanModel and CorrelationModel methods:
    • apply() renamed to value()
    • jac() renamed to jacobian()
  • Fix prediction computation when using linear regression (#52)
• ego:
  • Refactor Egor using argmin 0.7.0 solver framework EgorSolver can be used with argmin::Executor and benefit from observers and checkpointing features (#67)
  • Egor use kriging setting by default (i.e. one cluster with constant mean and squared exponential correlation model)
• Add notebook on Manuau Loa CO2 example to show GpMix/Gpx surrogate model usage (#62)
• Use xoshiro instead of isaac random generator (#63)
• Upgrade ndarray 0.15, linfa 0.6.1, PyO3 0.17 (#57, #64)

Version 0.6.0 - 2022-11-14

• gp: Kriging derivatives predictions are implemented (#44, #45), derivatives for Gp with linear regression are implemented (#47)
  • predict_derivatives: prediction of the output derivatives y wtr the input x
  • predict_variance_derivatives: prediction of the derivatives of the output variance wrt the input x
• moe: as for gp, derivatives methods for smooth and hard predictions are implemented (#46)
• ego: when available derivatives are used to optimize the infill criterion with slsqp (#44)
• egobox Python binding: add GpMix/Gpx in Python egobox module, the Python binding of egobox-moe::Moe (#31)

Version 0.5.0 - 2022-10-07

• Add Egor minimize interruption capability (Ctrl+C) from Python (#30)
• Minor performance improvement in moe clustering (#29)
• Improvements following JOSS submission review (#34, #36, #38, #39, #40, #42)

Version 0.4.0 - 2022-07-09

• Generate Python egobox module for Linux (#20)
• Improve Egor robustness by adding LHS optimization (#21)
• Improve moe with automatic number of clusters determination (#22)
• Use linfa 0.6.0 making BLAS dependency optional (#23)
• Improve Egor by implementing automatic reclustering every 10-points addition (#25)
• Fix Egor parallel infill strategy (qEI): bad objectives and constraints gp models updste (#26)

Version 0.3.0 - 2022-05-05

Improve mixture of experts (#15)

• Implement moe save/load (feature persistent)
• Rename GpSurrogate to Surrogate
• Remove fit_for_predict
• Implement ParamGuard for MoeParams
• Implement Fit for MoeParams
• Rename MoeParams setters

Refactor moe/ego relation (#16)

• Move MoeFit as SurrogateBuilder from moe to ego
• Implement SurrogateBuilder for Moe
• Moe uses linfa::Fit trait
• Rename Evaluator as PreProcessor

Refactor MixintEgor (#17)

• Rename PreProcessor::eval to run
• Implement linfa::Fit for MixintMoeParams, use linfa::Dataset
• Rename SurrogateParams to MoeBuilder
• Rename n_parallel to q_parallel (qEI stategy)

Version 0.2.1 - 2022-04-13

• Improve documentation
• egobox Python module: rename egobox Optimizer class to Egor

Version 0.2.0 - 2022-03-24

• Add hot start
• Add constraint handling
• Add mixed-integer optimization capability
• Add Python binding with PyO3

Version 0.1.0 - 2021-11-18

Initial release

• doe: LHS, FullFactorial, Random sampling
• gp: Gaussian Process models with 3 regression models (constant, linear quadratic) and 4 correlation models (squared exponential, absolute exponential, matern32, matern52)
• moe: Mixture of Experts: find the bests mix of gps given a number of clusters regarding smooth or hard recombination
• ego: Contains egor optimizer which is a super EGO algorithm implemented on top of the previous elements. It implements several infill strategy: EI, WB2, WB2S and use either COBYLA or SLSQP for internal optimization.