Relax IVF AQ FastScan (facebookresearch#2940)

Summary:
Pull Request resolved: facebookresearch#2940

This test fails on some occasions.
After investigation it turns out this is due to non reproducible behavior IndexIVFFastScan::search_implem_14 with a parallel loop, where there are ties in the resutls (ie. the resulting distances are the same but not the ids).
As a workaround I relaxed the test slightly.
+ a fix in the checksum function.

Reviewed By: algoriddle

Differential Revision: D47229086

fbshipit-source-id: 55e53bcfe47cf33041cc7fd5691b5de65067ce0f

faiss/python/extra_wrappers.py

@@ -108,7 +108,7 @@ def checksum(a):
  """ compute a checksum for quick-and-dirty comparisons of arrays """
  a = a.view('uint8')
  if a.ndim == 1:
- return bvec_checksum(s.size, swig_ptr(a))
+ return bvec_checksum(a.size, swig_ptr(a))
  n, d = a.shape
  cs = np.zeros(n, dtype='uint64')
  bvecs_checksum(n, d, swig_ptr(a), swig_ptr(cs))

tests/test_fast_scan_ivf.py

@@ -558,7 +558,7 @@ def subtest_accuracy(self, aq, st, by_residual, implem, metric_type='L2'):
  recall1 = (I1 == gt).sum() / nq
 
  print(aq, st, by_residual, implem, metric_type, recall_ref, recall1)
- assert abs(recall_ref - recall1) < 0.05
+ assert abs(recall_ref - recall1) < 0.051
 
  def xx_test_accuracy(self):
  # generated programatically below