Partly revert scikit-learn#6799 to keep DBSCAN fast (scikit-learn#9118)

qinhanmin2014 · ogrisel · commit c393bd499ed5 · 2017-06-15T10:45:24.000+02:00
diff --git a/sklearn/cluster/_dbscan_inner.pyx b/sklearn/cluster/_dbscan_inner.pyx
@@ -4,7 +4,6 @@
 
 cimport cython
 from libcpp.vector cimport vector
-from libcpp.set cimport set as cset
 cimport numpy as np
 import numpy as np
 
@@ -23,7 +22,6 @@ def dbscan_inner(np.ndarray[np.uint8_t, ndim=1, mode='c'] is_core,
     cdef np.npy_intp i, label_num = 0, v
     cdef np.ndarray[np.npy_intp, ndim=1] neighb
     cdef vector[np.npy_intp] stack
-    cdef cset[np.npy_intp] seen
 
     for i in range(labels.shape[0]):
         if labels[i] != -1 or not is_core[i]:
@@ -40,15 +38,12 @@ def dbscan_inner(np.ndarray[np.uint8_t, ndim=1, mode='c'] is_core,
                     neighb = neighborhoods[i]
                     for i in range(neighb.shape[0]):
                         v = neighb[i]
-                        if labels[v] == -1 and seen.count(v) == 0:
-                            seen.insert(v)
+                        if labels[v] == -1:
                             push(stack, v)
 
             if stack.size() == 0:
                 break
             i = stack.back()
             stack.pop_back()
 
-        seen.clear()
-
         label_num += 1
