Avoid duplicates in RandomSearch

Why:

The filtering of duplicates in RandomSearch is a lot more efficient than
relying on the database with DuplicateKeyErrors. There will still be
race conditions leading to this error, but filtering upfront in
RandomSearch reduces significantly the number of registration failures
and lead to better workload balance between the workers. Without this
fix, some workers end up always updating their algo and trying to sample
the next point but loose all race conditions.

How:

Use _trial_info to filter out duplicates while sampling points...

src/orion/algo/base.py

@@ -18,6 +18,10 @@
 log = logging.getLogger(__name__)
 
 
+def infer_trial_id(point):
+    return hashlib.md5(str(list(point)).encode('utf-8')).hexdigest()
+
+
 # pylint: disable=too-many-public-methods
 class BaseAlgorithm(object, metaclass=ABCMeta):
     """Base class describing what an algorithm can do.
@@ -187,11 +191,10 @@ def observe(self, points, results):
 
         """
         for point, result in zip(points, results):
-            _point = list(point)
-            _id = hashlib.md5(str(_point).encode('utf-8')).hexdigest()
+            point_id = infer_trial_id(point)
 
-            if _id not in self._trials_info:
-                self._trials_info[_id] = result
+            if point_id not in self._trials_info:
+                self._trials_info[point_id] = result
 
     @property
     def is_done(self):

src/orion/algo/random.py

@@ -10,7 +10,7 @@
 """
 import numpy
 
-from orion.algo.base import BaseAlgorithm
+from orion.algo.base import BaseAlgorithm, infer_trial_id
 
 
 class Random(BaseAlgorithm):
@@ -57,4 +57,15 @@ def suggest(self, num=1):
         .. note:: New parameters must be compliant with the problem's domain
            `orion.algo.space.Space`.
         """
-        return self.space.sample(num, seed=tuple(self.rng.randint(0, 1000000, size=3)))
+        points = []
+        point_ids = set(self._trials_info.keys())
+        i = 0
+        while len(points) < num:
+            new_point = self.space.sample(1, seed=tuple(self.rng.randint(0, 1000000, size=3)))[0]
+            point_id = infer_trial_id(new_point)
+            if point_id not in point_ids:
+                point_ids.add(point_id)
+                points.append(new_point)
+            i += 1
+
+        return points

tests/unittests/algo/test_random.py

@@ -43,3 +43,38 @@ def test_set_state(space):
 
     random_search.set_state(state)
     assert numpy.allclose(a, random_search.suggest(1)[0])
+
+
+def test_suggest_unique():
+    """Verify that RandomSearch do not sample duplicates"""
+    space = Space()
+    space.register(Integer('yolo1', 'uniform', -3, 6))
+
+    random_search = Random(space)
+
+    n_samples = 6
+    values = sum(random_search.suggest(n_samples), tuple())
+    assert len(values) == n_samples
+    assert len(set(values)) == n_samples
+
+
+def test_suggest_unique_history():
+    """Verify that RandomSearch do not sample duplicates based observed points"""
+    space = Space()
+    space.register(Integer('yolo1', 'uniform', -3, 6))
+
+    random_search = Random(space)
+
+    n_samples = 3
+    values = sum(random_search.suggest(n_samples), tuple())
+    assert len(values) == n_samples
+    assert len(set(values)) == n_samples
+
+    random_search.observe([[value] for value in values], [1] * n_samples)
+
+    n_samples = 3
+    new_values = sum(random_search.suggest(n_samples), tuple())
+    assert len(new_values) == n_samples
+    assert len(set(new_values)) == n_samples
+    # No duplicates
+    assert (set(new_values) & set(values)) == set()