Maintainance (#183)

* Remove derive_more

* Update CHANGELOG

* Increase trust region in case of EGO success

* Improve TREGO logging

* Test trego on rosenbrock 2d

* Fix documentation

* Increase maxiter to "ensure" optim test success

Author: relf

CHANGELOG.md

@@ -1,5 +1,25 @@
 # Changes
 
+## Version 0.21.1 - 31/07/2024
+
+* `gp`: Fix variance gradient computation by @relf in <https://github.com/relf/egobox/pull/177>
+
+## Version 0.21.0 - 09/07/2024
+
+* Implement [TREGO algorithm](https://arxiv.org/abs/2101.06808) by @relf in <https://github.com/relf/egobox/pull/173>
+* Fix added point count in TREGO local step by @relf in <https://github.com/relf/egobox/pull/174>
+* Fix WB2S criteria scaling factor and fmin computation by @relf in <https://github.com/relf/egobox/pull/175>
+* Prepare release 0.21 by @relf in <https://github.com/relf/egobox/pull/176>
+
+## Version 0.20.0 - 25/06/2024
+
+* Make `n_optmod` option available in Python by @relf in <https://github.com/relf/egobox/pull/161>
+* Add dependabot cargo ecosystem check by @relf in <https://github.com/relf/egobox/pull/163>
+* Save original parameters in trained `gp` models  by @relf in <https://github.com/relf/egobox/pull/166>
+* Implement cross validation metric for surrogates by @relf in <https://github.com/relf/egobox/pull/167>
+* Better `Egor` solver state handling by @relf in <https://github.com/relf/egobox/pull/168>
+* Refactor `ego` module by @relf in <https://github.com/relf/egobox/pull/169>
+
 ## Version 0.19.0 - 15/05/2024
 
 * `ego`:

README.md

@@ -195,7 +195,7 @@ Jones, D. R., Schonlau, M., & Welch, W. J. (1998). Efficient global optimization
 black-box functions. Journal of Global Optimization, 13(4), 455–492.
 
 Diouane, Youssef, et al. "TREGO: a trust-region framework for efficient global optimization."
-Journal of Global Optimization 86.1 (2023): 1-23.
+Journal of Global Optimization 86.1 (2023): 1-23. <https://arxiv.org/pdf/2101.06808>
 
 smtorg. (2018). Surrogate modeling toolbox. In GitHub repository. GitHub. <https://github.com/SMTOrg/smt>
 

ego/Cargo.toml

@@ -57,7 +57,6 @@ serde_json = "1"
 typetag = { version = "0.2" }
 dyn-clonable = { version = "0.9" }
 
-derive_more = "0.99"
 
 [dev-dependencies]
 criterion = "0.5"

ego/src/egor.rs

@@ -390,14 +390,14 @@ mod tests {
 
     #[test]
     #[serial]
-    fn test_rosenbrock_2d_infill_target_egor_builder() {
+    fn test_rosenbrock_2d_trego_egor_builder() {
         let now = Instant::now();
         let xlimits = array![[-2., 2.], [-2., 2.]];
         let doe = Lhs::new(&xlimits)
             .with_rng(Xoshiro256Plus::seed_from_u64(42))
             .sample(10);
         let res = EgorBuilder::optimize(rosenb)
-            .configure(|config| config.doe(&doe).max_iters(20).seed(42))
+            .configure(|config| config.doe(&doe).max_iters(20).seed(42).trego(true))
             .min_within(&xlimits)
             .run()
             .expect("Minimize failure");
@@ -445,7 +445,7 @@ mod tests {
                 config
                     .n_cstr(2)
                     .doe(&doe)
-                    .max_iters(30)
+                    .max_iters(50)
                     .cstr_tol(array![2e-6, 1e-6])
                     .seed(42)
             })

ego/src/lib.rs

@@ -38,7 +38,7 @@
 //! The implementation relies on [Mixture of Experts](egobox_moe).
 //!
 //!
-//! ## Mixed-integer optmization
+//! ## Mixed-integer optimization
 //!
 //! While [Egor] optimizer works with continuous data (i.e floats), the optimizer
 //! allows to make basic mixed-integer optimization. The configuration of the Optimizer
@@ -181,14 +181,15 @@
 //! (2019). [A python surrogate modeling framework with derivatives](https://doi.org/10.1016/j.advengsoft.2019.03.005).
 //! Advances in Engineering Software, 102662.
 //!
-//! Dubreuil, S., Bartoli, N., Gogu, C., & Lefebvre, T. (2020). (Towards an efficient global multi-
-//! disciplinary design optimization algorithm)[https://doi.org/10.1007/s00158-020-02514-6].
+//! Dubreuil, S., Bartoli, N., Gogu, C., & Lefebvre, T. (2020). [Towards an efficient global multi-
+//! disciplinary design optimization algorithm](https://doi.org/10.1007/s00158-020-02514-6).
 //! Structural and Multidisciplinary Optimization, 62(4), 1739–1765.
 //!
 //! Jones, D. R., Schonlau, M., & Welch, W. J. (1998). Efficient global optimization of expensive
 //! black-box functions. Journal of Global Optimization, 13(4), 455–492.
 //!
-//! \[<a id="Diouane2023">Diouane(2023)</a>\]: Diouane, Youssef, et al. "TREGO: a trust-region framework for efficient global optimization."
+//! \[<a id="Diouane2023">Diouane(2023)</a>\]: Diouane, Youssef, et al.
+//! [TREGO: a trust-region framework for efficient global optimization](https://arxiv.org/pdf/2101.06808)
 //! Journal of Global Optimization 86.1 (2023): 1-23.
 //!
 //! smtorg. (2018). Surrogate modeling toolbox. In [GitHub repository](https://github.com/SMTOrg/smt)

ego/src/solver/egor_solver.rs

@@ -326,14 +326,16 @@ where
         let prev_best = state.prev_best_index.unwrap(); // initialized in init
 
         // Check prev step success
+        let diff = y_data[[prev_best, 0]] - rho(state.sigma);
+        let last_iter_success = y_data[[best, 0]] < diff;
         info!(
-            "success = {} as {} < {} - {}",
-            y_data[[best, 0]] < (y_data[[prev_best, 0]] - rho(state.sigma)),
+            "success = {} as {} {} {} - {}",
+            last_iter_success,
             y_data[[best, 0]],
+            if last_iter_success { "<" } else { ">=" },
             y_data[[prev_best, 0]],
             rho(state.sigma)
         );
-        let last_iter_success = y_data[[best, 0]] < (y_data[[prev_best, 0]] - rho(state.sigma));
         let mut new_state = state.clone();
         if !state.prev_step_ego && state.get_iter() != 0 {
             // Adjust trust region wrt local step success
@@ -352,6 +354,18 @@ where
                     old, new_state.sigma
                 );
             }
+        } else if state.get_iter() != 0 {
+            // Adjust trust region wrt global step success
+            if last_iter_success {
+                let old = state.sigma;
+                new_state.sigma *= self.config.trego.gamma;
+                info!(
+                    "Previous EGO global step successful: sigma {} -> {}",
+                    old, new_state.sigma
+                );
+            } else {
+                info!("Previous EGO global step not successful");
+            }
         }
 
         let is_global_phase = (last_iter_success && state.prev_step_ego)