make temporary index copy in BruteForce

* needs more memory, but safer than in-place and simpler than workarounds

ann_benchmarks.py

@@ -178,20 +178,20 @@ def query(self, v, n):
 class BruteForce(BaseANN):
     """kNN search that uses a linear scan = brute force."""
     def __init__(self, metric):
-        assert metric in ('angular', ), "BruteForce doesn't support metric %s" % metric
+        if metric not in ('angular', ):
+            raise NotImplementedError("BruteForce doesn't support metric %s" % metric)
         self._metric = metric
         self.name = 'BruteForce()'
 
     def fit(self, X):
-        """Initialize the search index. Modifies `X` in place!"""
-        X /= np.sqrt((X ** 2).sum(-1))[..., np.newaxis]  # normalize vectors to unit length
-        self.index = X
+        """Initialize the search index."""
+        self.index = X / np.sqrt((X ** 2).sum(-1))[..., np.newaxis]  # normalize vectors to unit length
 
     def query(self, v, n):
         """Find indices of `n` most similar vector from index to query vector `v`."""
         v /= np.sqrt((v ** 2).sum())  # normalize query to unit length
-        cossims = numpy.dot(self.index, v)  # cossim = dot over normalized
-        indices = np.argsort(cossims)[::-1]  # sort by cossim
+        cossims = numpy.dot(self.index, v)  # cossim = dot product over normalized vectors
+        indices = np.argsort(cossims)[::-1]  # sort by cossim, highest first
         return indices[:n]  # return top N most similar