Relax IVFFlatDedup test (#3077)

Summary:
Pull Request resolved: #3077

This diff relaxes some IVFFlatDedup tests where distances are slighlty different over runs.
Should fix

https://app.circleci.com/pipelines/github/facebookresearch/faiss/4709/workflows/8c8213bf-8fe0-4c4e-9a7d-991f44bf1010/jobs/25551

https://app.circleci.com/pipelines/github/facebookresearch/faiss/4709/workflows/8c8213bf-8fe0-4c4e-9a7d-991f44bf1010/jobs/25547

Reviewed By: algoriddle

Differential Revision: D49732349

fbshipit-source-id: 728b9885c6b7d6ba697ccb6bacc0abd0ee2b0679

contrib/evaluation.py

@@ -226,23 +226,41 @@ def compute_PR_for(q):
 # Functions that compare search results with a reference result.
 # They are intended for use in tests
 
-def check_ref_knn_with_draws(Dref, Iref, Dnew, Inew):
-    """ test that knn search results are identical, raise if not """
-    np.testing.assert_array_almost_equal(Dref, Dnew, decimal=5)
+def _cluster_tables_with_tolerance(tab1, tab2, thr):
+    """ for two tables, cluster them by merging values closer than thr.
+    Returns the cluster ids for each table element """
+    tab = np.hstack([tab1, tab2])
+    tab.sort()
+    n = len(tab)
+    diffs = np.ones(n)
+    diffs[1:] = tab[1:] - tab[:-1]
+    unique_vals = tab[diffs > thr]
+    idx1 = np.searchsorted(unique_vals, tab1, side='right') - 1
+    idx2 = np.searchsorted(unique_vals, tab2, side='right') - 1
+    return idx1, idx2
+
+
+def check_ref_knn_with_draws(Dref, Iref, Dnew, Inew, rtol=1e-5):
+    """ test that knn search results are identical, with possible ties.
+    Raise if not. """
+    np.testing.assert_allclose(Dref, Dnew, rtol=rtol)
     # here we have to be careful because of draws
     testcase = unittest.TestCase()   # because it makes nice error messages
     for i in range(len(Iref)):
         if np.all(Iref[i] == Inew[i]): # easy case
             continue
-        # we can deduce nothing about the latest line
-        skip_dis = Dref[i, -1]
-        for dis in np.unique(Dref):
-            if dis == skip_dis:
+
+        # otherwise collect elements per distance
+        r = rtol * Dref[i].max()
+
+        DrefC, DnewC = _cluster_tables_with_tolerance(Dref[i], Dnew[i], r)
+
+        for dis in np.unique(DrefC):
+            if dis == DrefC[-1]:
                 continue
-            mask = Dref[i, :] == dis
+            mask = DrefC == dis
             testcase.assertEqual(set(Iref[i, mask]), set(Inew[i, mask]))
 
-
 def check_ref_range_results(Lref, Dref, Iref,
                             Lnew, Dnew, Inew):
     """ compare range search results wrt. a reference result,

tests/test_index_composite.py

@@ -17,7 +17,7 @@
 from common_faiss_tests import get_dataset_2
 from faiss.contrib.datasets import SyntheticDataset
 from faiss.contrib.inspect_tools import make_LinearTransform_matrix
-
+from faiss.contrib.evaluation import check_ref_knn_with_draws
 
 class TestRemoveFastScan(unittest.TestCase):
     def do_test(self, ntotal, removed):
@@ -430,12 +430,6 @@ def test_mmappedIO_pretrans(self):
 
 class TestIVFFlatDedup(unittest.TestCase):
 
-    def normalize_res(self, D, I):
-        dmax = D[-1]
-        res = [(d, i) for d, i in zip(D, I) if d < dmax]
-        res.sort()
-        return res
-
     def test_dedup(self):
         d = 10
         nb = 1000
@@ -471,10 +465,7 @@ def test_dedup(self):
         Dref, Iref = index_ref.search(xq, 20)
         Dnew, Inew = index_new.search(xq, 20)
 
-        for i in range(nq):
-            ref = self.normalize_res(Dref[i], Iref[i])
-            new = self.normalize_res(Dnew[i], Inew[i])
-            assert ref == new
+        check_ref_knn_with_draws(Dref, Iref, Dnew, Inew)
 
         # test I/O
         fd, tmpfile = tempfile.mkstemp()
@@ -487,10 +478,7 @@ def test_dedup(self):
                 os.unlink(tmpfile)
         Dst, Ist = index_st.search(xq, 20)
 
-        for i in range(nq):
-            new = self.normalize_res(Dnew[i], Inew[i])
-            st = self.normalize_res(Dst[i], Ist[i])
-            assert st == new
+        check_ref_knn_with_draws(Dnew, Inew, Dst, Ist)
 
         # test remove
         toremove = np.hstack((np.arange(3, 1000, 5), np.arange(850, 950)))
@@ -501,10 +489,7 @@ def test_dedup(self):
         Dref, Iref = index_ref.search(xq, 20)
         Dnew, Inew = index_new.search(xq, 20)
 
-        for i in range(nq):
-            ref = self.normalize_res(Dref[i], Iref[i])
-            new = self.normalize_res(Dnew[i], Inew[i])
-            assert ref == new
+        check_ref_knn_with_draws(Dref, Iref, Dnew, Inew)
 
 
 class TestSerialize(unittest.TestCase):