@@ -1,6 +1,9 @@

+import operator

+import pickle

 import random

 import cloudpickle

+import dill

 import pytest

 from adaptive.learner import (

@@ -24,22 +27,37 @@ def goal_2(learner):

     return learner.npoints >= 20

+learners_pairs = [

+    (Learner1D, dict(bounds=(-1, 1))),

+    (Learner2D, dict(bounds=[(-1, 1), (-1, 1)])),

+    (LearnerND, dict(bounds=[(-1, 1), (-1, 1), (-1, 1)])),

+    (SequenceLearner, dict(sequence=list(range(100)))),

+    (IntegratorLearner, dict(bounds=(0, 1), tol=1e-3)),

+    (AverageLearner, dict(atol=0.1)),

+]

+

+serializers = (pickle, dill, cloudpickle)

+

+learners = [

+    (learner_type, learner_kwargs, serializer)

+    for serializer in serializers

+    for learner_type, learner_kwargs in learners_pairs

+]

+

+

+def f_for_pickle_balancing_learner(x):

+    return 1

+

+

+def f_for_pickle_datasaver(x):

+    return dict(x=x, y=x)

+

+

 @pytest.mark.parametrize(

-    "learner_type, learner_kwargs",

-    [

-        (Learner1D, dict(bounds=(-1, 1))),

-        (Learner2D, dict(bounds=[(-1, 1), (-1, 1)])),

-        (LearnerND, dict(bounds=[(-1, 1), (-1, 1), (-1, 1)])),

-        (SequenceLearner, dict(sequence=list(range(100)))),

-        (IntegratorLearner, dict(bounds=(0, 1), tol=1e-3)),

-        (AverageLearner, dict(atol=0.1)),

-    ],

+    "learner_type, learner_kwargs, serializer", learners,

 )

-def test_cloudpickle_for(learner_type, learner_kwargs):

-    """Test serializing a learner using cloudpickle.

-

-    We use cloudpickle because with pickle the functions are only

-    pickled by reference."""

+def test_serialization_for(learner_type, learner_kwargs, serializer):

+    """Test serializing a learner using different serializers."""

     def f(x):

         return random.random()

@@ -49,9 +67,10 @@ def test_cloudpickle_for(learner_type, learner_kwargs):

     simple(learner, goal_1)

     learner_bytes = cloudpickle.dumps(learner)

-    # Delete references

-    del f

-    del learner

+    if serializer is not pickle:

+        # With pickle the functions are only pickled by reference

+        del f

+        del learner

     learner_loaded = cloudpickle.loads(learner_bytes)

     assert learner_loaded.npoints >= 10

@@ -59,45 +78,61 @@ def test_cloudpickle_for(learner_type, learner_kwargs):

     assert learner_loaded.npoints >= 20

-def test_cloudpickle_for_datasaver():

+@pytest.mark.parametrize(

+    "serializer", serializers,

+)

+def test_serialization_for_datasaver(serializer):

     def f(x):

         return dict(x=1, y=x ** 2)

+    if serializer is pickle:

+        # f from the local scope cannot be pickled

+        f = f_for_pickle_datasaver  # noqa: F811

+

     _learner = Learner1D(f, bounds=(-1, 1))

-    learner = DataSaver(_learner, arg_picker=lambda x: x["y"])

+    learner = DataSaver(_learner, arg_picker=operator.itemgetter("y"))

     simple(learner, goal_1)

-    learner_bytes = cloudpickle.dumps(learner)

+    learner_bytes = serializer.dumps(learner)

-    # Delete references

-    del f

-    del _learner

-    del learner

+    if serializer is not pickle:

+        # With pickle the functions are only pickled by reference

+        del f

+        del _learner

+        del learner

-    learner_loaded = cloudpickle.loads(learner_bytes)

+    learner_loaded = serializer.loads(learner_bytes)

     assert learner_loaded.npoints >= 10

     simple(learner_loaded, goal_2)

     assert learner_loaded.npoints >= 20

-def test_cloudpickle_for_balancing_learner():

+@pytest.mark.parametrize(

+    "serializer", serializers,

+)

+def test_serialization_for_balancing_learner(serializer):

     def f(x):

         return x ** 2

+    if serializer is pickle:

+        # f from the local scope cannot be pickled

+        f = f_for_pickle_balancing_learner  # noqa: F811

+

     learner_1 = Learner1D(f, bounds=(-1, 1))

     learner_2 = Learner1D(f, bounds=(-2, 2))

     learner = BalancingLearner([learner_1, learner_2])

     simple(learner, goal_1)

-    learner_bytes = cloudpickle.dumps(learner)

+    learner_bytes = serializer.dumps(learner)

-    # Delete references

-    del f

-    del learner_1

-    del learner_2

-    del learner

+    if serializer is not pickle:

+        # With pickle the functions are only pickled by reference

+        del f

+        del learner_1

+        del learner_2

+        del learner

-    learner_loaded = cloudpickle.loads(learner_bytes)

+    learner_loaded = serializer.loads(learner_bytes)

     assert learner_loaded.npoints >= 10

     simple(learner_loaded, goal_2)

     assert learner_loaded.npoints >= 20

@@ -44,6 +44,7 @@ extras_require = {

     ],

     "testing": [

         "cloudpickle",

+        "dill",

         "flaky",

         "pytest",

         "pytest-cov",

@@ -65,4 +65,4 @@ include_trailing_comma=True

 force_grid_wrap=0

 use_parentheses=True

 line_length=88

-known_third_party=PIL,atomicwrites,cloudpickle,flaky,holoviews,matplotlib,nbconvert,numpy,pytest,scipy,setuptools,skopt,sortedcollections,sortedcontainers

+known_third_party=PIL,atomicwrites,cloudpickle,dill,flaky,holoviews,matplotlib,nbconvert,numpy,pytest,scipy,setuptools,skopt,sortedcollections,sortedcontainers