Merge pull request #97 from MunchLab/update-matisse

add dist matrix function and update tutorial

Author: yemeen

doc_source/notebooks/Matisse/example_matisse.ipynb

@@ -1,6 +1,6 @@
 [project]
 name = "ect"
-version = "1.2.3"
+version = "1.2.4"
 authors = [
   { name="Liz Munch", email="muncheli@msu.edu" },
 ]

pyproject.toml

@@ -13,6 +13,7 @@
 from .ect import ECT
 from .embed_complex import EmbeddedComplex, EmbeddedGraph, EmbeddedCW
 from .directions import Directions
+from .results import ECTResult
 from .sect import SECT
 from .dect import DECT
 from .utils import examples
@@ -25,5 +26,6 @@
     "EmbeddedGraph",
     "EmbeddedCW",
     "Directions",
+    "ECTResult",
     "examples",
 ]

src/ect/__init__.py

@@ -1,7 +1,7 @@
 import matplotlib.pyplot as plt
 import numpy as np
 from ect.directions import Sampling
-from scipy.spatial.distance import cdist
+from scipy.spatial.distance import cdist, pdist, squareform
 from typing import Union, List, Callable
 
 
@@ -319,7 +319,7 @@ def _plot_ecc(self, theta):
     def dist(
         self,
         other: Union["ECTResult", List["ECTResult"]],
-        metric: Union[str, Callable] = "cityblock",
+        metric: Union[str, Callable] = "frobenius",
         **kwargs,
     ):
         """
@@ -365,7 +365,15 @@ def dist(
                     f"Shape mismatch at index {i}: {self.shape} vs {ect.shape}"
                 )
 
-        # use ravel to avoid copying the data and compute distances
+        if isinstance(metric, str) and metric.lower() in ("frobenius", "fro"):
+            a = np.asarray(self, dtype=np.float64)
+            if single:
+                b = np.asarray(other, dtype=np.float64)
+                return float(np.sqrt(np.sum((a - b) ** 2)))
+            b = np.stack([np.asarray(ect, dtype=np.float64) for ect in others], axis=0)
+            diff = b - a
+            return np.sqrt(np.sum(diff * diff, axis=(1, 2)))
+
         distances = cdist(
             self.ravel()[np.newaxis, :],
             np.vstack([ect.ravel() for ect in others]),
@@ -374,3 +382,30 @@ def dist(
         )[0]
 
         return distances[0] if single else distances
+
+    @classmethod
+    def dist_matrix(
+        cls,
+        results: List["ECTResult"],
+        metric: Union[str, Callable] = "frobenius",
+        **kwargs,
+    ) -> np.ndarray:
+        if not results:
+            return np.empty((0, 0), dtype=np.float64)
+
+        shape0 = results[0].shape
+        for i, r in enumerate(results):
+            if r.shape != shape0:
+                raise ValueError(f"Shape mismatch at index {i}: {shape0} vs {r.shape}")
+
+        if isinstance(metric, str) and metric.lower() in ("frobenius", "fro"):
+            return np.vstack([results[i].dist(results, metric="frobenius") for i in range(len(results))])
+
+        if isinstance(metric, str):
+            X = np.stack([np.asarray(r, dtype=np.float64).ravel() for r in results], axis=0)
+            try:
+                return squareform(pdist(X, metric=metric, **kwargs))
+            except TypeError:
+                return cdist(X, X, metric=metric, **kwargs)
+
+        return np.vstack([results[i].dist(results, metric=metric, **kwargs) for i in range(len(results))])

src/ect/results.py

@@ -90,11 +90,19 @@ def test_dist_single_ectresult(self):
         result2_modified.directions = result2.directions
         result2_modified.thresholds = result2.thresholds
 
-        # Test L1 distance (default)
-        dist_l1 = self.result.dist(result2_modified)
-        expected_l1 = np.abs(self.result - result2_modified).sum()
-        self.assertAlmostEqual(dist_l1, expected_l1)
-        self.assertIsInstance(dist_l1, (float, np.floating))
+        # Test frobenius distance (default)
+        dist_frobenius = self.result.dist(result2_modified)
+        expected_frobenius = np.sqrt(
+            np.sum(
+                (
+                    np.asarray(self.result, dtype=np.float64)
+                    - np.asarray(result2_modified, dtype=np.float64)
+                ).ravel()
+                ** 2
+            )
+        )
+        self.assertAlmostEqual(dist_frobenius, expected_frobenius)
+        self.assertIsInstance(dist_frobenius, (float, np.floating))
 
         # Test L2 distance
         dist_l2 = self.result.dist(result2_modified, metric="euclidean")
@@ -119,7 +127,7 @@ def test_dist_list_of_ectresults(self):
             r.thresholds = self.result.thresholds
 
         # Test batch distances
-        distances = self.result.dist([result2, result3, result4])
+        distances = self.result.dist([result2, result3, result4], metric="cityblock")
 
         # Check return type is array
         self.assertIsInstance(distances, np.ndarray)