vamos-optimization 1.2.0__tar.gz → 1.2.1__tar.gz

{vamos_optimization-1.2.0 → vamos_optimization-1.2.1}/CHANGELOG.md

@@ -4,6 +4,39 @@ All notable public changes to VAMOS are documented here.
 
 ## [Unreleased]
 
+## [1.2.1] - 2026-11-15
+
+VAMOS 1.2.1 is a patch release. It corrects C2-DTLZ2's default radius with two
+objectives, which changes results of affected experiments, and the constraint
+count of eight ZXH-CF problems, keeps persistent Optuna studies of the tuning
+CLI from mixing with those of earlier versions, and corrects two install
+instructions in the documentation. See the [1.2.1 release
+notes](docs/project/release-notes-1.2.1.md).
+
+### Fixed
+
+- C2-DTLZ2's default radius with two objectives is 0.5, as in Jain and Deb
+  (2014) and jMetal, instead of 0.2: the whole quarter circle of its front is
+  feasible, and its packaged two-objective front is regenerated. This changes
+  the results of two-objective C2-DTLZ2 runs; `r=0.2` gives the former problem
+  (#280).
+- ZXH-CF5–8 and ZXH-CF13–16 have one constraint per constrained angle, the
+  k_theta constraints of Zhou, Xiang and He (2021), as ZXH-CF9–12 do: `k + 1`
+  or `k + 2` constraints instead of `2k + 1` or `2k + 2`. Feasibility and the
+  overall constraint violation do not change; techniques that use the
+  individual constraints, such as `violated_constraints`, can rank differently
+  (#281).
+- `vamos tune` stores persistent Optuna studies under
+  `__vamos_cli_final_population_hv_v2`. VAMOS 1.2.0 changed the built-in search
+  spaces and scores but kept `_v1`, so resuming an earlier study mixed trials of
+  two contracts or stopped with Optuna's `ValueError` ("CategoricalDistribution
+  does not support dynamic value space"); such a study now starts afresh
+  (#282).
+- Documentation: the troubleshooting guide no longer lists the removed
+  `research` extra and gives notebooks' `[analysis,examples]` install, and the
+  algorithms reference describes the removed baselines with the 1.1.0 install
+  command.
+
 ## [1.2.0] - 2026-11-14
 
 VAMOS 1.2.0 corrects built-in algorithms, operators, problems, reference
@@ -606,7 +639,8 @@ development markers; they are not prior public releases.
 - Studio, tuning, visualization, provider integrations, plugin interfaces, and
   non-stable CLI commands remain experimental.
 
-[Unreleased]: https://github.com/vamos-optimization/VAMOS/compare/v1.2.0...HEAD
+[Unreleased]: https://github.com/vamos-optimization/VAMOS/compare/v1.2.1...HEAD
+[1.2.1]: https://github.com/vamos-optimization/VAMOS/releases/tag/v1.2.1
 [1.2.0]: https://github.com/vamos-optimization/VAMOS/releases/tag/v1.2.0
 [1.1.0]: https://github.com/vamos-optimization/VAMOS/releases/tag/v1.1.0
 [1.0.2]: https://github.com/vamos-optimization/VAMOS/releases/tag/v1.0.2

{vamos_optimization-1.2.0 → vamos_optimization-1.2.1}/CITATION.cff

@@ -17,6 +17,6 @@ authors:
   given-names: "J. Manuel"
   affiliation: "Universidad Rey Juan Carlos, Spain"
 title: "VAMOS: Vectorized Architecture for Multiobjective Optimization Studies"
-version: 1.2.0
-date-released: 2026-11-14
+version: 1.2.1
+date-released: 2026-11-15
 url: "https://github.com/vamos-optimization/VAMOS"

{vamos_optimization-1.2.0 → vamos_optimization-1.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vamos-optimization
-Version: 1.2.0
+Version: 1.2.1
 Summary: Vectorized Architecture for Multiobjective Optimization Studies
 Author: Nicolás R. Uribe, Alberto Herrán, Antonio J. Nebro, Javier Del Ser, J. Manuel Colmenar
 Maintainer: Nicolás R. Uribe
@@ -131,7 +131,9 @@ back to their published definitions, which changes results of affected
 experiments, and adds 115 constrained benchmark problems, jMetal's
 constraint-handling techniques and the CCMO, C-TAEA, ToP and CMOEA-MS
 algorithms. It supports Python 3.13 and 3.14 and is the last minor release for
-Python 3.10. See the
+Python 3.10. VAMOS 1.2.1 corrects C2-DTLZ2's default radius with two objectives
+and keeps persistent Optuna studies from mixing with earlier versions. See the
+[1.2.1 release notes](https://vamos-optimization.org/project/release-notes-1.2.1/), the
 [1.2.0 release notes](https://vamos-optimization.org/project/release-notes-1.2.0/), including
 which results to rerun, and the earlier
 [1.1.0 release notes](https://vamos-optimization.org/project/release-notes-1.1.0/).
@@ -206,7 +208,7 @@ pip install "smac>=2.0"
 | `numba` | Stable optional | Accelerates core numeric kernels such as variation, tournament selection, and MOEA/D neighborhood updates. |
 | `moocore` | Stable optional | Adds accelerated quality-indicator support, especially hypervolume-style metrics. |
 
-In VAMOS 1.2.0, use `engine="auto"` (CLI: `--engine auto`)
+In VAMOS 1.2.1, use `engine="auto"` (CLI: `--engine auto`)
 to prefer loadable Numba for all nine built-in algorithms, with a NumPy fallback.
 Numba remains optional. Omitting `engine` still selects NumPy, preserving the 1.x
 default; explicit `engine="numba"` fails if unavailable. First use can spend more
@@ -513,7 +515,7 @@ If you use VAMOS in published work, cite it directly:
   title = {VAMOS: Vectorized Architecture for Multiobjective Optimization Studies},
   author = {Rodriguez Uribe, Nicolas and Herr{\'a}n, Alberto and Nebro, Antonio J. and Del Ser, Javier and Colmenar, J. Manuel},
   year = {2026},
-  version = {1.2.0},
+  version = {1.2.1},
   url = {https://github.com/vamos-optimization/VAMOS}
 }
 ```

{vamos_optimization-1.2.0 → vamos_optimization-1.2.1}/README.md

@@ -17,7 +17,9 @@ back to their published definitions, which changes results of affected
 experiments, and adds 115 constrained benchmark problems, jMetal's
 constraint-handling techniques and the CCMO, C-TAEA, ToP and CMOEA-MS
 algorithms. It supports Python 3.13 and 3.14 and is the last minor release for
-Python 3.10. See the
+Python 3.10. VAMOS 1.2.1 corrects C2-DTLZ2's default radius with two objectives
+and keeps persistent Optuna studies from mixing with earlier versions. See the
+[1.2.1 release notes](https://vamos-optimization.org/project/release-notes-1.2.1/), the
 [1.2.0 release notes](https://vamos-optimization.org/project/release-notes-1.2.0/), including
 which results to rerun, and the earlier
 [1.1.0 release notes](https://vamos-optimization.org/project/release-notes-1.1.0/).
@@ -92,7 +94,7 @@ pip install "smac>=2.0"
 | `numba` | Stable optional | Accelerates core numeric kernels such as variation, tournament selection, and MOEA/D neighborhood updates. |
 | `moocore` | Stable optional | Adds accelerated quality-indicator support, especially hypervolume-style metrics. |
 
-In VAMOS 1.2.0, use `engine="auto"` (CLI: `--engine auto`)
+In VAMOS 1.2.1, use `engine="auto"` (CLI: `--engine auto`)
 to prefer loadable Numba for all nine built-in algorithms, with a NumPy fallback.
 Numba remains optional. Omitting `engine` still selects NumPy, preserving the 1.x
 default; explicit `engine="numba"` fails if unavailable. First use can spend more
@@ -399,7 +401,7 @@ If you use VAMOS in published work, cite it directly:
   title = {VAMOS: Vectorized Architecture for Multiobjective Optimization Studies},
   author = {Rodriguez Uribe, Nicolas and Herr{\'a}n, Alberto and Nebro, Antonio J. and Del Ser, Javier and Colmenar, J. Manuel},
   year = {2026},
-  version = {1.2.0},
+  version = {1.2.1},
   url = {https://github.com/vamos-optimization/VAMOS}
 }
 ```

{vamos_optimization-1.2.0 → vamos_optimization-1.2.1}/src/vamos/experiment/cli/_tune_runtime.py

@@ -33,7 +33,7 @@ MODEL_BACKENDS = ("optuna", "bohb_optuna", "smac3", "bohb", "irace")
 NON_MODEL_BACKENDS = ("racing", "random")
 ALL_BACKENDS = NON_MODEL_BACKENDS + MODEL_BACKENDS
 _ARCHIVE_TUNING_PARAMS = {"use_external_archive", "archive_unbounded", "archive_prune_policy"}
-_OPTUNA_CLI_STUDY_SUFFIX = "__vamos_cli_final_population_hv_v1"
+_OPTUNA_CLI_STUDY_SUFFIX = "__vamos_cli_final_population_hv_v2"
 # SMPSO is scored on its leaders archive instead of its final swarm, a separate scoring contract.
 _OPTUNA_CLI_LEADERS_SUFFIX = "__vamos_cli_leaders_hv_v1"
 _LEADER_SCORED_ALGORITHMS = ("smpso",)

{vamos_optimization-1.2.0 → vamos_optimization-1.2.1}/src/vamos/foundation/problem/constrained_many.py

@@ -161,12 +161,8 @@ class CDTLZProblem(Problem):
                 else:
                     r = 15.0
         elif key == "c2dtlz2" and r is None:
-            if self.n_obj == 2:
-                r = 0.2
-            elif self.n_obj == 3:
-                r = 0.4
-            else:
-                r = 0.5
+            # Jain and Deb (2014), as jMetal's C2_DTLZ2: 0.4 with three objectives and 0.5 otherwise.
+            r = 0.4 if self.n_obj == 3 else 0.5
         self._r = float(r) if r is not None else None
 
     def evaluate(self, X: np.ndarray[Any, Any], out: dict[str, np.ndarray[Any, Any]]) -> None:

{vamos_optimization-1.2.0 → vamos_optimization-1.2.1}/src/vamos/foundation/problem/zxh_cf.py

@@ -71,19 +71,21 @@ class ZXHCFProblem(Problem):
     """ZXH-CF``variant`` (1-16) with ``n_var`` variables in ``[1e-10, 1 - 1e-10]`` and ``n_obj`` objectives.
 
     Constraints follow VAMOS's convention, ``g <= 0`` feasible. They come first ``S + h - 1`` and, when
-    ``r > 0``, ``r - (S + h)``, which are jMetal's values with the opposite sign. Then, for each of the
-    first k angles:
+    ``r > 0``, ``r - (S + h)``, which are jMetal's values with the opposite sign. Then one constraint for
+    each of the first k angles, the paper's k_theta constraints ``v_j(theta) <= 0`` (its equation (6)):
 
-    - type B: ``theta - 3/4`` and ``1/4 - theta``, as in jMetal;
+    - type B: ``max(theta - 3/4, 1/4 - theta)``, positive only outside [1/4, 3/4];
     - type C: ``min(theta - 1/4, 3/4 - theta)``, positive only inside (1/4, 3/4);
-    - type D: ``min(theta - 1/10, 2/5 - theta)`` and ``min(theta - 7/10, 4/5 - theta)``.
-
-    A constraint of type C or D excludes an interval, so it is not a pair of constraints that must both
-    hold. jMetal stores the two bounds of each interval, ``<= 0`` feasible, and sets the overall violation
-    and the number of violated constraints itself: an interval adds the smaller of its two values when both
-    are positive. VAMOS stores that smaller value, which gives jMetal's feasibility and overall violation
-    with one column per interval, as PlatEMO does. jMetal counts two violated constraints for an interval of
-    type C and three for one of type D; VAMOS counts the column once.
+    - type D: ``max(min(theta - 1/10, 2/5 - theta), min(theta - 7/10, 4/5 - theta))``, positive only inside
+      (1/10, 2/5) or (7/10, 4/5).
+
+    jMetal, whose classes are the authors' Java code, stores both bounds of each angle of type B, ``>= 0``
+    feasible, and both bounds of each excluded interval of types C and D, positive when violated, with the
+    overall violation and the number of violated constraints precomputed: an interval adds the smaller of its
+    two values when both are positive. VAMOS's column per angle gives the same feasibility and overall
+    violation, because an angle can be outside only one side of [1/4, 3/4] and inside only one excluded
+    interval at a time; jMetal counts two violated constraints for an angle of type B or C and three for one
+    of type D, VAMOS one.
     """
 
     def __init__(self, variant: int, *, n_var: int = 13, n_obj: int = 3) -> None:
@@ -107,7 +109,7 @@ class ZXHCFProblem(Problem):
         self.constraint_type = _TYPES[(number - 1) // 4]
         self.radius = ((number - 1) % 4) / 4.0
         self.k = constrained_angles(n_obj)
-        per_angle = {"A": 0, "B": 2, "C": 1, "D": 2}[self.constraint_type]
+        per_angle = 0 if self.constraint_type == "A" else 1
         self.n_constraints = (2 if self.radius > 0.0 else 1) + per_angle * self.k
         self.xl = np.full(self.n_var, _MARGIN, dtype=float)
         self.xu = np.full(self.n_var, 1.0 - _MARGIN, dtype=float)
@@ -151,11 +153,12 @@ class ZXHCFProblem(Problem):
         for i in range(self.k):
             t = theta[:, i]
             if self.constraint_type == "B":
-                columns.extend([t - _BETA, -(t - _ALPHA)])
+                columns.append(np.maximum(t - _BETA, -(t - _ALPHA)))
             elif self.constraint_type == "C":
                 columns.append(np.minimum(t - _ALPHA, -(t - _BETA)))
             elif self.constraint_type == "D":
-                columns.extend(np.minimum(t - low, -(t - high)) for low, high in _D_INTERVALS)
+                (low1, high1), (low2, high2) = _D_INTERVALS
+                columns.append(np.maximum(np.minimum(t - low1, -(t - high1)), np.minimum(t - low2, -(t - high2))))
         return columns
 
     def evaluate(self, X: np.ndarray[Any, Any], out: dict[str, np.ndarray[Any, Any]]) -> None:

{vamos_optimization-1.2.0 → vamos_optimization-1.2.1}/src/vamos/foundation/version.py

@@ -5,7 +5,7 @@ Version helpers for VAMOS.
 from __future__ import annotations
 
 # Bumped by hand in the release pull request; see docs/release_smoke.md.
-__version__ = "1.2.0"
+__version__ = "1.2.1"
 
 
 def get_version() -> str: