[ENH] Add LIPOOptimizer backend integration (#102)

README.md

@@ -80,6 +80,7 @@ pip install hyperactive
 ```bash
 pip install hyperactive[sklearn-integration]  # scikit-learn integration
 pip install hyperactive[sktime-integration]   # sktime/skpro integration
+pip install hyperactive[lipo-integration]     # lipo global optimizer
 pip install hyperactive[all_extras]           # Everything including Optuna
 ```
 
@@ -111,7 +112,7 @@ pip install hyperactive[all_extras]           # Everything including Optuna
     </td>
     <td width="33%">
       <a href="https://hyperactive.readthedocs.io/en/latest/user_guide/optimizers/optuna.html"><b>Multiple Backends</b></a><br>
-      <sub>GFO algorithms, Optuna samplers, and sklearn search methods through one unified API.</sub>
+      <sub>GFO algorithms, Optuna samplers, sklearn search methods, and lipo's parameter-free global optimizer through one unified API.</sub>
     </td>
     <td width="33%">
       <a href="https://hyperactive.readthedocs.io/en/latest/api_reference.html"><b>Stable & Tested</b></a><br>
@@ -177,13 +178,13 @@ flowchart TB
             GFO["GFO<br/>21 algorithms"]
             OPTUNA["Optuna<br/>8 algorithms"]
             SKL["sklearn<br/>2 algorithms"]
-            MORE["...<br/>more to come"]
+            LIPO["LIPO<br/>1 algorithm"]
         end
 
         OPT --> GFO
         OPT --> OPTUNA
         OPT --> SKL
-        OPT --> MORE
+        OPT --> LIPO
     end
 
     subgraph OUT["Output"]
@@ -366,6 +367,34 @@ best_params = optimizer.solve()
 
 
 
+<details>
+<summary><b>LIPO Global Optimizer</b></summary>
+
+```python
+import numpy as np
+from hyperactive.opt.lipo import LIPOOptimizer
+
+def objective(params):
+    x, y = params["x"], params["y"]
+    return -(x**2 + y**2)
+
+search_space = {
+    "x": np.arange(-5, 5, 0.1),
+    "y": np.arange(-5, 5, 0.1),
+}
+
+optimizer = LIPOOptimizer(
+    search_space=search_space,
+    n_iter=100,
+    experiment=objective,
+)
+best_params = optimizer.solve()
+```
+
+</details>
+
+
+
 <details>
 <summary><b>Time Series Forecasting with sktime</b></summary>
 
@@ -508,4 +537,4 @@ If you use this software in your research, please cite:
 
 ## License
 
-[MIT License](./LICENSE) - Free for commercial and academic use.
+[MIT License](./LICENSE) - Free for commercial and academic use.

examples/lipo/lipo_examples.py

@@ -0,0 +1,18 @@
+import numpy as np
+from hyperactive.opt.lipo import LIPOOptimizer
+
+
+def objective(params):
+    x, y = params["x"], params["y"]
+    return -(x ** 2 + y ** 2)   # max at (0, 0)
+
+
+opt = LIPOOptimizer(
+    search_space={
+        "x": np.arange(-5, 5, 0.1),
+        "y": np.arange(-5, 5, 0.1),
+    },
+    n_iter=100,
+    experiment=objective,
+)
+print(opt.solve())   # {'x': ~0.0, 'y': ~0.0}

pyproject.toml

@@ -99,6 +99,7 @@ all_extras = [
   "optuna<5",
   "cmaes",  # Required for CmaEsOptimizer (optuna's CMA-ES sampler)
   "lightning",
+  "lipo",
 ]
 
 

src/hyperactive/opt/__init__.py

@@ -4,6 +4,7 @@
 
 from hyperactive.opt.gridsearch import GridSearchSk
 from hyperactive.opt.random_search import RandomSearchSk
+from .lipo import LIPOOptimizer
 
 from .gfo import (
     BayesianOptimizer,

src/hyperactive/opt/lipo.py

@@ -0,0 +1,54 @@
+import numpy as np
+from lipo import GlobalOptimizer
+
+class LIPOOptimizer:
+    """Parameter-free global optimizer via the lipo package."""
+
+    def __init__(self, search_space, n_iter, experiment, maximize=True):
+        self.search_space = search_space
+        self.n_iter = n_iter
+        self.experiment = experiment
+        self.maximize = maximize
+
+    def _parse_search_space(self):
+        lower, upper, cats = {}, {}, {}
+        for key, values in self.search_space.items():
+            # Categorical: list of strings
+            if isinstance(values, list) and isinstance(values[0], str):
+                cats[key] = values
+            else:
+                arr = np.array(values)
+                lower[key] = float(arr.min())
+                upper[key] = float(arr.max())
+                # Store grid so we can snap results back later
+                self._grids = getattr(self, "_grids", {})
+                self._grids[key] = arr
+        return lower, upper, cats
+
+    def _snap_to_grid(self, params):
+        """Snap lipo's continuous output to nearest valid grid point."""
+        snapped = {}
+        for key, val in params.items():
+            if key in getattr(self, "_grids", {}):
+                grid = self._grids[key]
+                snapped[key] = grid[np.argmin(np.abs(grid - val))]
+            else:
+                snapped[key] = val   # categorical, pass through
+        return snapped
+
+    def solve(self):
+        lower, upper, cats = self._parse_search_space()
+
+        def wrapped(**kwargs):
+            return self.experiment(self._snap_to_grid(kwargs))
+
+        opt = GlobalOptimizer(
+            wrapped,
+            lower_bounds=lower,
+            upper_bounds=upper,
+            categories=cats,
+            maximize=self.maximize,
+        )
+        opt.run(self.n_iter)
+
+        return self._snap_to_grid(opt.optimum[0])

src/hyperactive/tests/test_lipo.py

@@ -0,0 +1,41 @@
+import numpy as np
+import pytest
+from hyperactive.opt.lipo import LIPOOptimizer
+
+
+def sphere(params):
+    return -(params["x"] ** 2 + params["y"] ** 2)
+
+
+def test_lipo_basic():
+    opt = LIPOOptimizer(
+        search_space={
+            "x": np.arange(-5, 5, 0.1),
+            "y": np.arange(-5, 5, 0.1),
+        },
+        n_iter=50,
+        experiment=sphere,
+    )
+    best = opt.solve()
+    assert "x" in best and "y" in best
+    assert abs(best["x"]) < 1.5   # should be near 0
+
+
+def test_lipo_categorical():
+    def fn(p): return 1.0 if p["kernel"] == "rbf" else 0.0
+    opt = LIPOOptimizer(
+        search_space={"kernel": ["linear", "rbf", "poly"]},
+        n_iter=20, experiment=fn,
+    )
+    best = opt.solve()
+    assert best["kernel"] == "rbf"
+
+
+def test_lipo_snap_to_grid():
+    def fn(p): return -abs(p["x"] - 3)
+    opt = LIPOOptimizer(
+        search_space={"x": np.array([1, 2, 3, 4, 5])},
+        n_iter=30, experiment=fn,
+    )
+    best = opt.solve()
+    assert best["x"] in [1, 2, 3, 4, 5]  # must be on the grid