test: update tests with fixtures (#318)

.gitignore

@@ -2,6 +2,7 @@
 pytest/
 *.pytest_cache/
 .coverage
+.converage.*
 pytest-coverage.txt
 wandb/
 

tests/algos/__snapshots__/test_algos.ambr

@@ -77,17 +77,17 @@
   )
 # ---
 # name: TestTorchAlgos.test_reproducibility[stable_learning_control.algos.pytorch.sac.sac].1
-  -0.3054710924625397
+  -0.17646123468875885
 # ---
 # name: TestTorchAlgos.test_reproducibility[stable_learning_control.algos.pytorch.sac.sac].2
-  -0.5438500046730042
+  -0.7749449014663696
 # ---
 # name: TestTorchAlgos.test_reproducibility[stable_learning_control.algos.pytorch.sac.sac].3
-  -0.5551561713218689
+  -0.4696856439113617
 # ---
 # name: TestTorchAlgos.test_reproducibility[stable_learning_control.algos.pytorch.sac.sac].4
-  0.7831009030342102
+  0.8471362590789795
 # ---
 # name: TestTorchAlgos.test_reproducibility[stable_learning_control.algos.pytorch.sac.sac].5
-  -0.43937113881111145
+  -0.3496301770210266
 # ---

tests/algos/gpu/__snapshots__/test_algos_gpu.ambr

@@ -77,17 +77,17 @@
   )
 # ---
 # name: TestTorchAlgosGPU.test_reproducibility[gpu-stable_learning_control.algos.pytorch.sac.sac].1
-  -0.7098895311355591
+  -0.5978871583938599
 # ---
 # name: TestTorchAlgosGPU.test_reproducibility[gpu-stable_learning_control.algos.pytorch.sac.sac].2
-  -0.1639314442873001
+  -0.613029956817627
 # ---
 # name: TestTorchAlgosGPU.test_reproducibility[gpu-stable_learning_control.algos.pytorch.sac.sac].3
-  -0.7671857476234436
+  -0.7007529735565186
 # ---
 # name: TestTorchAlgosGPU.test_reproducibility[gpu-stable_learning_control.algos.pytorch.sac.sac].4
-  0.415353924036026
+  0.7336636781692505
 # ---
 # name: TestTorchAlgosGPU.test_reproducibility[gpu-stable_learning_control.algos.pytorch.sac.sac].5
-  -0.8201737403869629
+  -0.8068016171455383
 # ---

tests/algos/gpu/test_algos_gpu.py

@@ -20,21 +20,26 @@
 @pytest.mark.parametrize("algo", ALGOS)
 @pytest.mark.parametrize("device", ["gpu"])
 class TestTorchAlgosGPU:
-    env = gym.make("Pendulum-v1")  # Used because it is a simple environment.
+    @pytest.fixture
+    def env(self):
+        """Create Pendulum environment."""
+        env = gym.make("Pendulum-v1")  # Used because it is a simple environment.
 
-    # Seed the environment.
-    env.np_random, seed = seeding.np_random(0)
-    env.action_space.seed(0)
-    env.observation_space.seed(0)
+        # Seed the environment.
+        env.np_random, _ = seeding.np_random(0)
+        env.action_space.seed(0)
+        env.observation_space.seed(0)
 
-    def test_reproducibility(self, algo, device, snapshot):
+        return env
+
+    def test_reproducibility(self, algo, device, snapshot, env):
         """Checks if the algorithm is still working as expected."""
         # Import the algorithm run function.
         run = getattr(importlib.import_module(algo), algo.split(".")[-1])
 
         # Run the algorithm.
         trained_policy, replay_buffer = run(
-            lambda: self.env,
+            lambda: env,
             seed=0,
             epochs=1,
             update_after=400,
@@ -48,5 +53,5 @@ def test_reproducibility(self, algo, device, snapshot):
 
         # Test if the actions returned by the policy are the same.
         for _ in range(5):
-            action = trained_policy.get_action(self.env.observation_space.sample())
+            action = trained_policy.get_action(env.observation_space.sample())
             assert action == snapshot

tests/algos/test_algos.py

@@ -19,21 +19,26 @@
 
 @pytest.mark.parametrize("algo", ALGOS)
 class TestTorchAlgos:
-    env = gym.make("Pendulum-v1")  # Used because it is a simple environment.
+    @pytest.fixture
+    def env(self):
+        """Create Pendulum environment."""
+        env = gym.make("Pendulum-v1")  # Used because it is a simple environment.
 
-    # Seed the environment.
-    env.np_random, seed = seeding.np_random(0)
-    env.action_space.seed(0)
-    env.observation_space.seed(0)
+        # Seed the environment.
+        env.np_random, _ = seeding.np_random(0)
+        env.action_space.seed(0)
+        env.observation_space.seed(0)
 
-    def test_reproducibility(self, algo, snapshot):
+        return env
+
+    def test_reproducibility(self, algo, snapshot, env):
         """Checks if the algorithm is still working as expected."""
         # Import the algorithm run function.
         run = getattr(importlib.import_module(algo), algo.split(".")[-1])
 
         # Run the algorithm.
         trained_policy, replay_buffer = run(
-            lambda: self.env,
+            lambda: env,
             seed=0,
             epochs=1,
             update_after=400,
@@ -47,5 +52,5 @@ def test_reproducibility(self, algo, snapshot):
 
         # Test if the actions returned by the policy are the same.
         for _ in range(5):
-            action = trained_policy.get_action(self.env.observation_space.sample())
+            action = trained_policy.get_action(env.observation_space.sample())
             assert action == snapshot

tests/algos/tf2/gpu/test_tf2_algos_gpu.py

@@ -21,14 +21,19 @@
 @pytest.mark.parametrize("algo", ALGOS)
 @pytest.mark.parametrize("device", ["gpu"])
 class TestTF2AlgosGPU:
-    env = gym.make("Pendulum-v1")  # Used because it is a simple environment.
+    @pytest.fixture
+    def env(self):
+        """Create Pendulum environment."""
+        env = gym.make("Pendulum-v1")  # Used because it is a simple environment.
 
-    # Seed the environment.
-    env.np_random, seed = seeding.np_random(0)
-    env.action_space.seed(0)
-    env.observation_space.seed(0)
+        # Seed the environment.
+        env.np_random, _ = seeding.np_random(0)
+        env.action_space.seed(0)
+        env.observation_space.seed(0)
 
-    def test_reproducibility(self, algo, device, snapshot):
+        return env
+
+    def test_reproducibility(self, algo, device, snapshot, env):
         """Checks if the algorithm is still working as expected."""
         # Check if TensorFlow is available.
         if not importlib.util.find_spec("tensorflow"):
@@ -45,7 +50,7 @@ def test_reproducibility(self, algo, device, snapshot):
 
         # Run the algorithm.
         trained_policy, replay_buffer = run(
-            lambda: self.env,
+            lambda: env,
             seed=0,
             epochs=1,
             update_after=400,
@@ -59,5 +64,5 @@ def test_reproducibility(self, algo, device, snapshot):
 
         # Test if the actions returned by the policy are the same.
         for _ in range(5):
-            action = trained_policy.get_action(self.env.observation_space.sample())
+            action = trained_policy.get_action(env.observation_space.sample())
             assert action.numpy() == snapshot

tests/algos/tf2/test_tf2_algos.py

@@ -19,14 +19,19 @@
 
 @pytest.mark.parametrize("algo", ALGOS)
 class TestTF2Algos:
-    env = gym.make("Pendulum-v1")  # Used because it is a simple environment.
+    @pytest.fixture
+    def env(self):
+        """Create Pendulum environment."""
+        env = gym.make("Pendulum-v1")  # Used because it is a simple environment.
 
-    # Seed the environment.
-    env.np_random, seed = seeding.np_random(0)
-    env.action_space.seed(0)
-    env.observation_space.seed(0)
+        # Seed the environment.
+        env.np_random, _ = seeding.np_random(0)
+        env.action_space.seed(0)
+        env.observation_space.seed(0)
 
-    def test_reproducibility(self, algo, snapshot):
+        return env
+
+    def test_reproducibility(self, algo, snapshot, env):
         """Checks if the algorithm is still working as expected."""
         # Check if TensorFlow is available.
         if not importlib.util.find_spec("tensorflow"):
@@ -39,7 +44,7 @@ def test_reproducibility(self, algo, snapshot):
 
         # Run the algorithm.
         trained_policy, replay_buffer = run(
-            lambda: self.env,
+            lambda: env,
             seed=0,
             epochs=1,
             update_after=400,
@@ -53,5 +58,5 @@ def test_reproducibility(self, algo, snapshot):
 
         # Test if the actions returned by the policy are the same.
         for _ in range(5):
-            action = trained_policy.get_action(self.env.observation_space.sample())
+            action = trained_policy.get_action(env.observation_space.sample())
             assert action.numpy() == snapshot