Deployed 4d9e490 to 5.0 with MkDocs 1.6.0 and mike 2.1.2

Author: SimonBlanke

5.0/api/base_experiment/index.html

@@ -2192,7 +2192,14 @@ <h4 id="scoreparams"><code>score(params)</code></h4>
 <h4 id="paramnames"><code>paramnames()</code></h4>
 <p>Get the parameter names for the search space.</p>
 <p><strong>Returns:</strong>
-- <code>list</code>: List of parameter names</p>
+- <code>list[str] | None</code>: Parameter names. If <code>None</code>, any parameter keys are accepted.</p>
+<p>Notes:
+- When <code>paramnames()</code> returns a list, <code>params</code> passed to <code>evaluate/score</code> must be a subset of that list; otherwise a <code>ValueError</code> is raised.
+- <code>__call__(params)</code> is provided as shorthand for <code>score(params)</code> and returns only the score (float).
+- Score sign follows the tag <code>property:higher_or_lower_is_better</code>:
+  - <code>"higher"</code>: score equals evaluate (higher-is-better)
+  - <code>"lower"</code>: score is <code>-evaluate</code> (converted to higher-is-better)
+  - <code>"mixed"</code>: not supported by default; override <code>score</code> for custom behavior</p>
 <h3 id="tags-system">Tags System</h3>
 <p>BaseExperiment uses tags to specify optimization properties:</p>
 <h3 id="abstract-methods">Abstract Methods</h3>

5.0/api/optimizers/index.html

@@ -2415,7 +2415,7 @@ <h3 id="optuna-backend">Optuna Backend</h3>
 <p>The Optuna backend provides 8 modern optimization algorithms with advanced features like multi-objective optimization.</p>
 <p><strong>Import from:</strong> <code>hyperactive.opt.optuna</code> (also re-exported in <code>hyperactive.opt</code>)</p>
 <h3 id="scikit-learn-backend">Scikit-learn Backend</h3>
-<p>Direct integration with scikit-learn's optimization tools for familiar workflows.</p>
+<p>Sklearn-style search using <code>ParameterGrid</code> / <code>ParameterSampler</code>, with evaluation handled by Hyperactive experiments.</p>
 <p><strong>Import from:</strong> <code>hyperactive.opt.gridsearch</code> and <code>hyperactive.opt</code></p>
 <h2 id="gradient-free-optimizers-backend">Gradient-Free-Optimizers Backend</h2>
 <h3 id="hill-climbing-variants">Hill Climbing Variants</h3>
@@ -2507,11 +2507,11 @@ <h4 id="nsgaiiioptimizer">NSGAIIIOptimizer</h4>
 <h3 id="qmcoptimizer">QMCOptimizer</h3>
 <p>Quasi-Monte Carlo sampling.</p>
 <h2 id="scikit-learn-backend_1">Scikit-learn Backend</h2>
-<p>Direct integration with scikit-learn optimization tools.</p>
+<p>Sklearn-style search; not thin wrappers over sklearn CV utilities. Evaluation is performed by a <code>SklearnCvExperiment</code> and parallelization by Hyperactive backends.</p>
 <h3 id="gridsearchsk">GridSearchSk</h3>
-<p>Direct sklearn GridSearchCV integration.</p>
+<p>Sklearn-style exhaustive grid evaluation (evaluation via <code>SklearnCvExperiment</code>).</p>
 <h3 id="randomsearchsk">RandomSearchSk</h3>
-<p>Direct sklearn RandomizedSearchCV integration.</p>
+<p>Sklearn-style randomized search (evaluation via <code>SklearnCvExperiment</code>).</p>
 <h2 id="choosing-the-right-algorithm">Choosing the Right Algorithm</h2>
 <h3 id="for-quick-prototyping">For Quick Prototyping</h3>
 <ul>

5.0/experiments/custom_experiments/index.html

@@ -1951,6 +1951,13 @@ <h2 id="installation">Installation</h2>
 <p>The most recent version of Hyperactive is available on PyPi:</p>
 <div class="highlight"><pre><span></span><code><a id="__codelineno-0-1" name="__codelineno-0-1" href="#__codelineno-0-1"></a><span class="go">pip install hyperactive</span>
 </code></pre></div>
+<p>Optional integrations and backends:</p>
+<ul>
+<li>Optuna adapters: <code>pip install optuna&lt;5</code></li>
+<li>Sktime integrations: <code>pip install hyperactive[sktime-integration]</code></li>
+<li>Parallel backends (as needed): <code>pip install joblib dask ray</code></li>
+<li>Note: ray support depends on your Python version; see <code>pyproject.toml</code> constraints.</li>
+</ul>
 <h2 id="quick-start">Quick Start</h2>
 <h2 id="key-features_1">Key Features</h2>
 <h3 id="complete-architecture-design">Complete Architecture Design</h3>

5.0/index.html

@@ -1285,9 +1285,9 @@
       <ul class="md-nav__list">
 
           <li class="md-nav__item">
-  <a href="#experiment" class="md-nav__link">
+  <a href="#common-via-base-optuna-adapter" class="md-nav__link">
     <span class="md-ellipsis">
-      experiment
+      Common (via Base Optuna Adapter)
     </span>
   </a>
 
@@ -1303,9 +1303,18 @@
 </li>
 
           <li class="md-nav__item">
-  <a href="#acquisition_func" class="md-nav__link">
+  <a href="#deterministic_objective" class="md-nav__link">
     <span class="md-ellipsis">
-      acquisition_func
+      deterministic_objective
+    </span>
+  </a>
+  
+</li>
+        
+          <li class="md-nav__item">
+  <a href="#random_state" class="md-nav__link">
+    <span class="md-ellipsis">
+      random_state
     </span>
   </a>
 
@@ -2139,9 +2148,9 @@
       <ul class="md-nav__list">
 
           <li class="md-nav__item">
-  <a href="#experiment" class="md-nav__link">
+  <a href="#common-via-base-optuna-adapter" class="md-nav__link">
     <span class="md-ellipsis">
-      experiment
+      Common (via Base Optuna Adapter)
     </span>
   </a>
 
@@ -2157,9 +2166,18 @@
 </li>
 
           <li class="md-nav__item">
-  <a href="#acquisition_func" class="md-nav__link">
+  <a href="#deterministic_objective" class="md-nav__link">
+    <span class="md-ellipsis">
+      deterministic_objective
+    </span>
+  </a>
+  
+</li>
+        
+          <li class="md-nav__item">
+  <a href="#random_state" class="md-nav__link">
     <span class="md-ellipsis">
-      acquisition_func
+      random_state
     </span>
   </a>
 
@@ -2385,23 +2403,32 @@ <h2 id="about-the-implementation">About the Implementation</h2>
 - <strong>Principled exploration</strong>: Uses uncertainty to guide search
 - <strong>Non-parametric</strong>: Adapts to complex function shapes</p>
 <h2 id="parameters">Parameters</h2>
-<h3 id="experiment"><code>experiment</code></h3>
+<h3 id="common-via-base-optuna-adapter">Common (via Base Optuna Adapter)</h3>
 <ul>
-<li><strong>Type</strong>: <code>BaseExperiment</code></li>
-<li><strong>Description</strong>: The experiment object defining the optimization problem</li>
+<li><code>param_space</code> (dict): parameter space; tuples/lists treated as ranges/choices</li>
+<li><code>n_trials</code> (int): number of trials to run</li>
+<li><code>initialize</code> (dict | None): optional warm start/grid/vertices/random init</li>
+<li><code>early_stopping</code> (int | None): stop if no improvement after N trials</li>
+<li><code>max_score</code> (float | None): stop when reaching threshold</li>
+<li><code>experiment</code> (BaseExperiment): the experiment to optimize</li>
 </ul>
 <h3 id="n_startup_trials"><code>n_startup_trials</code></h3>
 <ul>
 <li><strong>Type</strong>: <code>int</code></li>
 <li><strong>Default</strong>: <code>10</code></li>
 <li><strong>Description</strong>: Number of random trials before GP optimization starts</li>
 </ul>
-<h3 id="acquisition_func"><code>acquisition_func</code></h3>
+<h3 id="deterministic_objective"><code>deterministic_objective</code></h3>
 <ul>
-<li><strong>Type</strong>: <code>str</code></li>
-<li><strong>Default</strong>: <code>"ei"</code> (Expected Improvement)</li>
-<li><strong>Options</strong>: <code>"ei"</code>, <code>"lcb"</code> (Lower Confidence Bound), <code>"pi"</code> (Probability of Improvement)</li>
-<li><strong>Description</strong>: Acquisition function for selecting next evaluation point</li>
+<li><strong>Type</strong>: <code>bool</code></li>
+<li><strong>Default</strong>: <code>False</code></li>
+<li><strong>Description</strong>: Whether the objective function is deterministic (passes through to Optuna's GPSampler).</li>
+</ul>
+<h3 id="random_state"><code>random_state</code></h3>
+<ul>
+<li><strong>Type</strong>: <code>int | None</code></li>
+<li><strong>Default</strong>: <code>None</code></li>
+<li><strong>Description</strong>: Seed for reproducibility (sets <code>seed</code> in the underlying GPSampler).</li>
 </ul>
 <h2 id="usage-example">Usage Example</h2>
 <h2 id="when-to-use-gp-optimizer">When to Use GP Optimizer</h2>
@@ -2417,24 +2444,13 @@ <h2 id="when-to-use-gp-optimizer">When to Use GP Optimizer</h2>
 - <strong>Very noisy objectives</strong>: More robust methods might be needed
 - <strong>Cheap evaluations</strong>: Random search might be sufficient</p>
 <h2 id="acquisition-functions">Acquisition Functions</h2>
+<p>Note: The specific acquisition function is handled internally by the Optuna sampler used by this adapter and is not user-configurable via this API. The following concepts are provided for background only.</p>
 <h3 id="expected-improvement-ei">Expected Improvement (EI)</h3>
-<ul>
-<li><strong>Best for</strong>: Balanced exploration-exploitation</li>
-<li><strong>Formula</strong>: <span class="arithmatex">\(EI(x) = \sigma(x) \cdot \phi(Z) + (\mu(x) - f_{best}) \cdot \Phi(Z)\)</span></li>
-<li><strong>Use when</strong>: General-purpose optimization</li>
-</ul>
+<p>Balanced exploration-exploitation using improvement probability and magnitude.</p>
 <h3 id="lower-confidence-bound-lcb">Lower Confidence Bound (LCB)</h3>
-<ul>
-<li><strong>Best for</strong>: Conservative optimization with uncertainty consideration</li>
-<li><strong>Formula</strong>: <span class="arithmatex">\(LCB(x) = \mu(x) - \kappa \cdot \sigma(x)\)</span></li>
-<li><strong>Use when</strong>: You want to avoid risky evaluations</li>
-</ul>
+<p>Conservative trade-off between mean prediction and uncertainty.</p>
 <h3 id="probability-of-improvement-pi">Probability of Improvement (PI)</h3>
-<ul>
-<li><strong>Best for</strong>: When you want high probability of improvement</li>
-<li><strong>Formula</strong>: <span class="arithmatex">\(PI(x) = \Phi(\frac{\mu(x) - f_{best}}{\sigma(x)})\)</span></li>
-<li><strong>Use when</strong>: Conservative improvement is preferred</li>
-</ul>
+<p>Focuses on points with high probability of improving over current best.</p>
 <h2 id="advanced-usage">Advanced Usage</h2>
 <h3 id="custom-acquisition-function">Custom Acquisition Function</h3>
 <h3 id="startup-trials-tuning">Startup Trials Tuning</h3>

5.0/optimizers/optuna/gp_optimizer/index.html

@@ -1222,9 +1222,9 @@
       <ul class="md-nav__list">
 
           <li class="md-nav__item">
-  <a href="#experiment" class="md-nav__link">
+  <a href="#common-via-base-optuna-adapter" class="md-nav__link">
     <span class="md-ellipsis">
-      experiment
+      Common (via Base Optuna Adapter)
     </span>
   </a>
 
@@ -1249,18 +1249,18 @@
 </li>
 
           <li class="md-nav__item">
-  <a href="#gamma" class="md-nav__link">
+  <a href="#weights" class="md-nav__link">
     <span class="md-ellipsis">
-      gamma
+      weights
     </span>
   </a>
 
 </li>
 
           <li class="md-nav__item">
-  <a href="#prior_weight" class="md-nav__link">
+  <a href="#random_state" class="md-nav__link">
     <span class="md-ellipsis">
-      prior_weight
+      random_state
     </span>
   </a>
 
@@ -2046,9 +2046,9 @@
       <ul class="md-nav__list">
 
           <li class="md-nav__item">
-  <a href="#experiment" class="md-nav__link">
+  <a href="#common-via-base-optuna-adapter" class="md-nav__link">
     <span class="md-ellipsis">
-      experiment
+      Common (via Base Optuna Adapter)
     </span>
   </a>
 
@@ -2073,18 +2073,18 @@
 </li>
 
           <li class="md-nav__item">
-  <a href="#gamma" class="md-nav__link">
+  <a href="#weights" class="md-nav__link">
     <span class="md-ellipsis">
-      gamma
+      weights
     </span>
   </a>
 
 </li>
 
           <li class="md-nav__item">
-  <a href="#prior_weight" class="md-nav__link">
+  <a href="#random_state" class="md-nav__link">
     <span class="md-ellipsis">
-      prior_weight
+      random_state
     </span>
   </a>
 
@@ -2199,10 +2199,14 @@ <h2 id="about-the-implementation">About the Implementation</h2>
 - <strong>Robust performance</strong>: Works well across many different optimization problems
 - <strong>Sample efficient</strong>: Requires fewer evaluations than random search</p>
 <h2 id="parameters">Parameters</h2>
-<h3 id="experiment"><code>experiment</code></h3>
+<h3 id="common-via-base-optuna-adapter">Common (via Base Optuna Adapter)</h3>
 <ul>
-<li><strong>Type</strong>: <code>BaseExperiment</code></li>
-<li><strong>Description</strong>: The experiment object defining the optimization problem</li>
+<li><code>param_space</code> (dict): parameter space; tuples/lists treated as ranges/choices</li>
+<li><code>n_trials</code> (int): number of trials to run</li>
+<li><code>initialize</code> (dict | None): optional warm start/grid/vertices/random init</li>
+<li><code>early_stopping</code> (int | None): stop if no improvement after N trials</li>
+<li><code>max_score</code> (float | None): stop when reaching threshold</li>
+<li><code>experiment</code> (BaseExperiment): the experiment to optimize</li>
 </ul>
 <h3 id="n_startup_trials"><code>n_startup_trials</code></h3>
 <ul>
@@ -2216,17 +2220,17 @@ <h3 id="n_ei_candidates"><code>n_ei_candidates</code></h3>
 <li><strong>Default</strong>: <code>24</code></li>
 <li><strong>Description</strong>: Number of candidate points to evaluate when computing Expected Improvement</li>
 </ul>
-<h3 id="gamma"><code>gamma</code></h3>
+<h3 id="weights"><code>weights</code></h3>
 <ul>
-<li><strong>Type</strong>: <code>float</code></li>
-<li><strong>Default</strong>: <code>0.25</code></li>
-<li><strong>Description</strong>: Quantile for splitting observations into good/bad groups. Lower values create more selective "good" groups.</li>
+<li><strong>Type</strong>: <code>callable | None</code></li>
+<li><strong>Default</strong>: <code>None</code></li>
+<li><strong>Description</strong>: Optional weighting function passed to Optuna's TPESampler.</li>
 </ul>
-<h3 id="prior_weight"><code>prior_weight</code></h3>
+<h3 id="random_state"><code>random_state</code></h3>
 <ul>
-<li><strong>Type</strong>: <code>float</code></li>
-<li><strong>Default</strong>: <code>1.0</code></li>
-<li><strong>Description</strong>: Prior weight for the Parzen estimators</li>
+<li><strong>Type</strong>: <code>int | None</code></li>
+<li><strong>Default</strong>: <code>None</code></li>
+<li><strong>Description</strong>: Seed for reproducibility (sets <code>seed</code> in the underlying TPESampler).</li>
 </ul>
 <h2 id="usage-example">Usage Example</h2>
 <h2 id="when-to-use-tpe">When to Use TPE</h2>

5.0/optimizers/optuna/tpe_optimizer/index.html

@@ -1827,14 +1827,15 @@
 
 
 <h1 id="scikit-learnstyled-optimizers">Scikit-learn–Styled Optimizers</h1>
-<p>Hyperactive provides scikit-learn–styled optimizers that use sklearn-style parameter grids and distributions while running through Hyperactive’s optimization and experiment abstractions. These are designed for users who prefer sklearn-like configuration but want to stay within the Hyperactive v5 architecture.</p>
+<p>Hyperactive provides sklearn-style search using <code>ParameterGrid</code> / <code>ParameterSampler</code>, with evaluation routed through Hyperactive experiments. This preserves familiar configuration while keeping the v5 architecture’s separation between search (optimizer) and evaluation (experiment).</p>
 <p>Key characteristics:</p>
 <ul>
 <li>Sklearn-style search spaces via <code>ParameterGrid</code> / <code>ParameterSampler</code></li>
 <li>Evaluation via a Hyperactive <code>Experiment</code> (typically <code>SklearnCvExperiment</code>)</li>
-<li>Parallelism via Hyperactive backends (<code>backend</code>, <code>backend_params</code>)</li>
+<li>Parallelism via optimizer backends (<code>backend</code>, <code>backend_params</code>)</li>
 <li>Results exposed on the optimizer (<code>best_params_</code>, <code>best_score_</code>, etc.)</li>
 </ul>
+<p>Note: These are not thin wrappers over <code>GridSearchCV</code> / <code>RandomizedSearchCV</code>. Cross‑validation is handled by the experiment (<code>SklearnCvExperiment</code>), and parallelization is controlled by Hyperactive’s backends.</p>