From b2999787e73d308fbba6877fda531f0a1ac2e1cd Mon Sep 17 00:00:00 2001
From: Jakub Adamczyk <jakubadamczyk10@gmail.com>
Date: Sun, 10 Aug 2025 13:56:23 +0200
Subject: [PATCH] Add GETAWAY feature names

---
 skfp/fingerprints/getaway.py  | 74 +++++++++++++++++++++++++++++++++++
 tests/fingerprints/getaway.py | 39 ++++++++++++++++++
 2 files changed, 113 insertions(+)

diff --git a/skfp/fingerprints/getaway.py b/skfp/fingerprints/getaway.py
index 9538778a..03de9c92 100644
--- a/skfp/fingerprints/getaway.py
+++ b/skfp/fingerprints/getaway.py
@@ -144,6 +144,80 @@ def __init__(
         )
         self.clip_val = clip_val
 
+    def get_feature_names_out(self, input_features=None) -> np.ndarray:  # noqa: ARG002
+        """
+        Get fingerprint output feature names. They correspond to various
+        descriptors derived from weighted Molecular Influence Matrix (MIM).
+        Definitions are complex and long, see references given in main body
+        for explanations, particularly Todeschini & Consonni book.
+
+        Parameters
+        ----------
+        input_features : array-like of str or None, default=None
+            Unused, kept for scikit-learn compatibility.
+
+        Returns
+        -------
+        feature_names_out : ndarray of str objects
+            GETAWAY feature names.
+        """
+        weighting_variants = [
+            "unweighted",
+            "atomic mass",
+            "van der Waals volume",
+            "electronegativity",
+            "polarizability",
+            "ion polarity",
+            "IState",
+        ]
+
+        feature_names = [
+            "total information content on the leverage equality (ITH)",
+            "standardized information content on the leverage equality (ISH)",
+            "mean information content on the leverage magnitude (HIC)",
+            "geometric mean of the leverage magnitude (HGM)",
+        ]
+
+        for weighting in weighting_variants:
+            h_indices = [f"{weighting} H index radius {radius}" for radius in range(9)]
+            total_h_index = f"{weighting} total H index"
+            feature_names.extend(h_indices)
+            feature_names.append(total_h_index)
+
+            hats_indices = [
+                f"{weighting} H matrix autocorrelation (HATS) radius {radius}"
+                for radius in range(9)
+            ]
+            total_hats_index = f"{weighting} total HATS index"
+            feature_names.extend(hats_indices)
+            feature_names.append(total_hats_index)
+
+        feature_names.extend(
+            [
+                "R-connectivity index (RCON)",
+                "average row sum of the influence/distance matrix (RARS",
+                "R-matrix leading eigenvalue (REIG)",
+            ]
+        )
+
+        for weighting in weighting_variants:
+            r_indices = [
+                f"{weighting} R index radius {radius}" for radius in range(1, 9)
+            ]
+            total_r_index = f"{weighting} R total index"
+            feature_names.extend(r_indices)
+            feature_names.append(total_r_index)
+
+            maximal_r_indices = [
+                f"{weighting} maximal R index (R+) radius {radius}"
+                for radius in range(1, 9)
+            ]
+            total_maximal_r_index = f"{weighting} maximal R total index"
+            feature_names.extend(maximal_r_indices)
+            feature_names.append(total_maximal_r_index)
+
+        return np.asarray(feature_names, dtype=object)
+
     def transform(
         self, X: Sequence[str | Mol], copy: bool = False
     ) -> np.ndarray | csr_array:
diff --git a/tests/fingerprints/getaway.py b/tests/fingerprints/getaway.py
index d6f0ba7f..df573fa9 100644
--- a/tests/fingerprints/getaway.py
+++ b/tests/fingerprints/getaway.py
@@ -17,3 +17,42 @@ def test_getaway_sparse_fingerprint(mols_conformers_list):
 
     assert X_skfp.shape == (len(mols_conformers_list), 273)
     assert np.issubdtype(X_skfp.dtype, np.floating)
+
+
+def test_getaway_feature_names():
+    getaway_fp = GETAWAYFingerprint()
+    feature_names = getaway_fp.get_feature_names_out()
+
+    assert len(feature_names) == getaway_fp.n_features_out
+    assert len(feature_names) == len(set(feature_names))
+
+    # for indices understanding, see:
+    # https://github.com/rdkit/rdkit/blob/53a01430e057d34cf7a7ca52eb2d8b069178114d/Code/GraphMol/Descriptors/GETAWAY.cpp#L1145
+    # and slide 28 from https://github.com/rdkit/UGM_2017/blob/master/Presentations/Godin_3D_Descriptors.pdf
+    # note that here we index from zero
+
+    assert "ITH" in feature_names[0]
+    assert "ISH" in feature_names[1]
+    assert "HIC" in feature_names[2]
+    assert "HGM" in feature_names[3]
+
+    assert all(
+        "total H index" in feature_names[idx] for idx in [13, 33, 53, 73, 93, 113, 133]
+    )
+    assert all(
+        "total HATS index" in feature_names[idx]
+        for idx in [23, 43, 63, 83, 103, 123, 143]
+    )
+
+    assert "RCON" in feature_names[144]
+    assert "RARS" in feature_names[145]
+    assert "REIG" in feature_names[146]
+
+    assert all(
+        "R total index" in feature_names[idx]
+        for idx in [155, 173, 191, 209, 227, 254, 263]
+    )
+    assert all(
+        "maximal R total index" in feature_names[idx]
+        for idx in [164, 182, 200, 218, 236, 254, 272]
+    )