hdgpso 0.1.0__py3-none-any.whl

hdgpso/__init__.py

@@ -0,0 +1,64 @@
+"""hdgpso package. Hybrid DE-GWO-PSO hyperparameter optimization.
+
+The :class:`HDGPSO` class is the optimizer described in the paper. It
+runs three operator stages on the same population in every iteration:
+Differential Evolution, Grey Wolf Optimization, and Particle Swarm
+Optimization. A RandomForest surrogate is fit on the trial history
+and is used to filter proposed candidates between the stages, using a
+tree-variance lower-confidence-bound score.
+
+Quick start::
+
+    from hdgpso import HDGPSO, SearchSpace, Float, Int, Categorical
+
+    space = SearchSpace({
+        "lr":     Float(1e-5, 1e-2, log=True),
+        "layers": Int(2, 8),
+        "act":    Categorical(["relu", "gelu", "tanh"]),
+    })
+
+    def objective(params):
+        # train a model with the given hyperparameters
+        return -val_accuracy   # lower is better
+
+    result = HDGPSO(space, objective, population_size=10,
+                    iterations=15, seed=0).optimize()
+    print(result.best_params, result.best_loss)
+
+The rank-based statistical helpers used for the paper benchmark
+(Friedman, Nemenyi, CD diagrams, Cliff's delta, bootstrap confidence
+intervals) are available in :mod:`hdgpso.stats`.
+
+.. note::
+
+   :class:`HDGPSOMF` is an experimental multi-fidelity variant built on
+   top of HDGPSO with BOHB-style successive halving. It is included for
+   future-work exploration only. It is not part of the published
+   results, and its API is not yet stable.
+"""
+from .core import (
+    Categorical,
+    Dimension,
+    Float,
+    HDGPSO,
+    Int,
+    OptimizeResult,
+    SearchSpace,
+)
+from .multifidelity import HDGPSOMF
+from ._version import __version__
+
+__all__ = [
+    # Search space
+    "Categorical",
+    "Dimension",
+    "Float",
+    "Int",
+    "SearchSpace",
+    # Optimizers
+    "HDGPSO",
+    "HDGPSOMF",
+    "OptimizeResult",
+    # Version
+    "__version__",
+]

hdgpso/_version.py

@@ -0,0 +1,2 @@
+"""Package version. Single source of truth — pyproject.toml reads it dynamically."""
+__version__ = "0.1.0"