get rid of the poorly architected list_distance option

flexible_clustering/fishdbc.py

@@ -63,7 +63,7 @@ class FISHDBC:
     """Flexible Incremental Scalable Hierarchical Density-Based Clustering."""
 
     def __init__(self, d, min_samples=5, m=5, ef=200, m0=None, level_mult=None,
-                 heuristic=True, balanced_add=True, list_distance=False):
+                 heuristic=True, balanced_add=True):
         """Setup the algorithm. The only mandatory parameter is d, the
         dissimilarity function. min_samples is passed to hdbscan, and
         the other parameters are all passed to HNSW."""
@@ -92,41 +92,15 @@ def __init__(self, d, min_samples=5, m=5, ef=200, m0=None, level_mult=None,
         self._distance_cache = distance_cache = {}
 
         # decorated_d will cache the computed distances in distance_cache.
-        if not list_distance:  # d is defined to work on scalars
-            def decorated_d(i, j):
-                # assert i == len(data) - 1 # 1st argument is the new item
-                if j in distance_cache:
-                    return distance_cache[j]
-                distance_cache[j] = dist = d(data[i], data[j])
-                return dist
-        if list_distance: # d is defined to work on a scalar and a list
-            def decorated_d(i, js):
-                assert i == len(data) - 1 # 1st argument is the new item
-                known = []
-                unknown_j, unknown_items = [], []
-                for pos, j in enumerate(js):
-                    k = j in distance_cache
-                    known.append(k)
-                    if not k:
-                        unknown_j.append(j)
-                        unknown_items.append(k)
-                new_d = d(data[i], unknown_items)
-                for j, dist in zip(unknown_j, new_d):
-                    distance_cache[j] = dist
-                old_d = (distance_cache[j] for j, k in zip(js, known) if k)
-                new_d = iter(new_d)
-
-                res = []
-                for k in known:
-                    if k:
-                        res.append(next(old_d))
-                    else:
-                        res.append(next(new_d))
-                return res
+        def decorated_d(i, j):
+            # assert i == len(data) - 1 # 1st argument is the new item
+            if j in distance_cache:
+                return distance_cache[j]
+            distance_cache[j] = dist = d(data[i], data[j])
+            return dist
 
         # We create the HNSW
-        the_hnsw = hnsw.HNSW(decorated_d, m, ef, m0, level_mult, heuristic,
-                             list_distance)
+        the_hnsw = hnsw.HNSW(decorated_d, m, ef, m0, level_mult, heuristic)
         self._hnsw_add = (the_hnsw.balanced_add if balanced_add
                           else the_hnsw.add)
 

flexible_clustering/hnsw.py

@@ -60,23 +60,14 @@ class HNSW(object):
     # where j is a neighbor of i and dist is distance
 
     def __init__(self, d, m=5, ef=200, m0=None, level_mult=None,
-                 heuristic=True, list_distance=False):
+                 heuristic=True):
         """d the dissimilarity function
 
         See other parameters in http://arxiv.org/pdf/1603.09320v2.pdf"""
 
         self.data = []
 
-        if list_distance:
-            def d1(x, y):
-                return d(x, [y])[0]
-            self.distance = d1
-            self.list_distance = d
-        else:
-            self.distance = d
-            def vd(x, ys):
-                return [d(x, y) for y in ys]
-            self.list_distance = vd
+        self.distance = d
 
         self._m = m
         self._ef = ef
@@ -219,7 +210,7 @@ def search(self, q, k=None, ef=None):
     def _search_graph_ef1(self, q, entry, dist, g):
         """Equivalent to _search_graph when ef=1."""
 
-        ld = self.list_distance
+        d = self.distance
         data = self.data
 
         best = entry
@@ -231,10 +222,11 @@ def _search_graph_ef1(self, q, entry, dist, g):
             dist, c = heappop(candidates)
             if dist > best_dist:
                 break
-            edges = [e for e in g[c] if e not in visited]
-            visited.update(edges)
-            dists = ld(q, [data[e] for e in edges])
-            for e, dist in zip(edges, dists):
+            for e in g[c]:
+                if e in visited:
+                    continue
+                visited.add(e)
+                dist = d(q, data[e])
                 if dist < best_dist:
                     best = e
                     best_dist = dist
@@ -245,7 +237,7 @@ def _search_graph_ef1(self, q, entry, dist, g):
 
     def _search_graph(self, q, ep, g, ef):
 
-        ld = self.list_distance
+        d = self.distance
         data = self.data
 
         candidates = [(-mdist, p) for mdist, p in ep]
@@ -259,10 +251,11 @@ def _search_graph(self, q, ep, g, ef):
             if dist > ref:
                 break
 
-            edges = [e for e in g[c] if e not in visited]
-            visited.update(edges)
-            dists = ld(q, [data[e] for e in edges])
-            for e, dist in zip(edges, dists):
+            for e in g[c]:
+                if e in visited:
+                    continue
+                visited.add(e)
+                dist = d(q, data[e])
                 mdist = -dist
                 if len(ep) < ef:
                     heappush(candidates, (dist, e))