reducers: add KernelPCA, Isomap, MDS, SpectralEmbedding, GaussianRandomProjection

app/web/main.py

@@ -210,6 +210,67 @@ REDUCERS: Dict[str, Dict[str, Any]] = {
             ("rotation", "str", "", ["", "varimax", "quartimax"], "Empty = None."),
         ],
     },
+    "sklearn.decomposition.KernelPCA": {
+        "pkg": "sklearn",
+        "label": "KernelPCA",
+        "blurb": "Non-linear PCA via the kernel trick. Deterministic; kernel choice shapes the output.",
+        "key": [
+            ("n_components", "int", 2, None, "Locked."),
+            ("kernel", "str", "rbf", ["linear", "poly", "rbf", "sigmoid", "cosine"], None),
+            ("random_state", "int", 42, None, None),
+        ],
+        "advanced": [
+            ("gamma", "str_or_float", "", None, "Empty = 1/n_features."),
+            ("degree", "int", 3, None, None),
+            ("coef0", "float", 1.0, None, None),
+            ("alpha", "float", 1.0, None, None),
+        ],
+    },
+    "sklearn.manifold.Isomap": {
+        "pkg": "sklearn",
+        "label": "Isomap",
+        "blurb": "Geodesic-distance manifold learning via shortest paths on a k-NN graph.",
+        "key": [
+            ("n_components", "int", 2, None, "Locked."),
+            ("n_neighbors", "int", 5, None, None),
+        ],
+        "advanced": [
+            ("metric", "str", "minkowski", None, None),
+            ("p", "int", 2, None, "Minkowski power (1 = manhattan, 2 = euclidean)."),
+            ("path_method", "str", "auto", ["auto", "FW", "D"], "Floyd-Warshall / Dijkstra / auto."),
+            ("neighbors_algorithm", "str", "auto", ["auto", "ball_tree", "kd_tree", "brute"], None),
+        ],
+    },
+    "sklearn.manifold.MDS": {
+        "pkg": "sklearn",
+        "label": "MDS",
+        "blurb": "Multidimensional scaling. Preserves pairwise distances; O(n²) memory.",
+        "key": [
+            ("n_components", "int", 2, None, "Locked."),
+            ("n_init", "int", 4, None, None),
+            ("random_state", "int", 42, None, None),
+        ],
+        "advanced": [
+            ("max_iter", "int", 300, None, None),
+            ("metric_mds", "bool", True, None, "Metric (True) vs non-metric MDS."),
+            ("metric", "str", "euclidean", None, None),
+            ("eps", "float", 1e-6, None, "Convergence tolerance."),
+        ],
+    },
+    "sklearn.manifold.SpectralEmbedding": {
+        "pkg": "sklearn",
+        "label": "SpectralEmbedding",
+        "blurb": "Laplacian eigenmaps on an affinity graph. What UMAP uses for initialisation.",
+        "key": [
+            ("n_components", "int", 2, None, "Locked."),
+            ("affinity", "str", "nearest_neighbors", ["nearest_neighbors", "rbf"], None),
+            ("random_state", "int", 42, None, None),
+        ],
+        "advanced": [
+            ("n_neighbors", "int_or_null", "", None, "For affinity=nearest_neighbors. Empty = n/10."),
+            ("gamma", "str_or_float", "", None, "For affinity=rbf. Empty = 1/n_features."),
+        ],
+    },
     "sklearn.manifold.TSNE": {
         "pkg": "sklearn",
         "label": "t-SNE",
@@ -295,6 +356,16 @@ REDUCERS: Dict[str, Dict[str, Any]] = {
             ("weight_adj", "float", 500.0, None, None),
         ],
     },
+    "sklearn.random_projection.GaussianRandomProjection": {
+        "pkg": "sklearn",
+        "label": "GaussianRandomProjection",
+        "blurb": "Johnson-Lindenstrauss baseline. Cheap, distance-preserving in expectation, structure-agnostic.",
+        "key": [
+            ("n_components", "int", 2, None, "Locked."),
+            ("random_state", "int", 42, None, None),
+        ],
+        "advanced": [],
+    },
 }