Annotate some functions that return None

Summary: Test functions return None. This codemod fixes that so type annotation efforts can focus on trickier cases.

Reviewed By: azad-meta

Differential Revision: D52570248

fbshipit-source-id: b20e5ec6cde1132d4e1f954af1e012d8464343c8

examples/mnist_lightning.py

@@ -136,7 +136,7 @@ def test_step(self, batch, batch_idx):
         self.log("test_accuracy", self.test_accuracy, on_step=False, on_epoch=True)
         return loss
 
-    def on_train_epoch_end(self):
+    def on_train_epoch_end(self) -> None:
         # Logging privacy spent: (epsilon, delta)
         epsilon = self.privacy_engine.get_epsilon(self.delta)
         self.log("epsilon", epsilon, on_epoch=True, prog_bar=True)

opacus/tests/accountants_test.py

@@ -27,7 +27,7 @@
 
 
 class AccountingTest(unittest.TestCase):
-    def test_rdp_accountant(self):
+    def test_rdp_accountant(self) -> None:
         noise_multiplier = 1.5
         sample_rate = 0.04
         steps = int(90 / 0.04)
@@ -39,7 +39,7 @@ def test_rdp_accountant(self):
         epsilon = accountant.get_epsilon(delta=1e-5)
         self.assertAlmostEqual(epsilon, 7.32911117143)
 
-    def test_gdp_accountant(self):
+    def test_gdp_accountant(self) -> None:
         noise_multiplier = 1.5
         sample_rate = 0.04
         steps = int(90 // 0.04)
@@ -52,7 +52,7 @@ def test_gdp_accountant(self):
         self.assertLess(6.59, epsilon)
         self.assertLess(epsilon, 6.6)
 
-    def test_prv_accountant(self):
+    def test_prv_accountant(self) -> None:
         noise_multiplier = 1.5
         sample_rate = 0.04
         steps = int(90 // 0.04)
@@ -65,7 +65,7 @@ def test_prv_accountant(self):
         epsilon = accountant.get_epsilon(delta=1e-5)
         self.assertAlmostEqual(epsilon, 6.777395712150674)
 
-    def test_get_noise_multiplier_rdp_epochs(self):
+    def test_get_noise_multiplier_rdp_epochs(self) -> None:
         delta = 1e-5
         sample_rate = 0.04
         epsilon = 8
@@ -81,7 +81,7 @@ def test_get_noise_multiplier_rdp_epochs(self):
 
         self.assertAlmostEqual(noise_multiplier, 1.416, places=4)
 
-    def test_get_noise_multiplier_rdp_steps(self):
+    def test_get_noise_multiplier_rdp_steps(self) -> None:
         delta = 1e-5
         sample_rate = 0.04
         epsilon = 8
@@ -96,7 +96,7 @@ def test_get_noise_multiplier_rdp_steps(self):
 
         self.assertAlmostEqual(noise_multiplier, 1.3562, places=4)
 
-    def test_get_noise_multiplier_prv_epochs(self):
+    def test_get_noise_multiplier_prv_epochs(self) -> None:
         delta = 1e-5
         sample_rate = 0.04
         epsilon = 8
@@ -112,7 +112,7 @@ def test_get_noise_multiplier_prv_epochs(self):
 
         self.assertAlmostEqual(noise_multiplier, 1.34765625, places=4)
 
-    def test_get_noise_multiplier_prv_steps(self):
+    def test_get_noise_multiplier_prv_steps(self) -> None:
         delta = 1e-5
         sample_rate = 0.04
         epsilon = 8
@@ -153,7 +153,7 @@ def test_get_noise_multiplier_overshoot(self, epsilon, epochs, sample_rate, delt
         actual_epsilon = accountant.get_epsilon(delta=delta)
         self.assertLess(actual_epsilon, epsilon)
 
-    def test_get_noise_multiplier_gdp(self):
+    def test_get_noise_multiplier_gdp(self) -> None:
         delta = 1e-5
         sample_rate = 0.04
         epsilon = 8
@@ -169,7 +169,7 @@ def test_get_noise_multiplier_gdp(self):
 
         self.assertAlmostEqual(noise_multiplier, 1.3232421875)
 
-    def test_accountant_state_dict(self):
+    def test_accountant_state_dict(self) -> None:
         noise_multiplier = 1.5
         sample_rate = 0.04
         steps = int(90 / 0.04)
@@ -191,7 +191,7 @@ def test_accountant_state_dict(self):
             accountant.state_dict(dummy_dest)["dummy_k"], dummy_dest["dummy_k"]
         )
 
-    def test_accountant_load_state_dict(self):
+    def test_accountant_load_state_dict(self) -> None:
         noise_multiplier = 1.5
         sample_rate = 0.04
         steps = int(90 / 0.04)

opacus/tests/batch_memory_manager_test.py

@@ -171,7 +171,7 @@ def test_empty_batch(
                     )
                     weights_before = torch.clone(model._module.fc.weight)
 
-    def test_equivalent_to_one_batch(self):
+    def test_equivalent_to_one_batch(self) -> None:
         torch.manual_seed(1337)
         model, optimizer, data_loader = self._init_training()
 
@@ -229,7 +229,7 @@ def test_equivalent_to_one_batch(self):
 class BatchMemoryManagerTestWithExpandedWeights(BatchMemoryManagerTest):
     GSM_MODE = "ew"
 
-    def test_empty_batch(self):
+    def test_empty_batch(self) -> None:
         pass
 
 

opacus/tests/ddp_hook_check.py

@@ -255,15 +255,15 @@ def run_function(local_function, tensor, dp, noise_multiplier=0, max_grad_norm=1
 
 
 class GradientComputationTest(unittest.TestCase):
-    def test_connection(self):
+    def test_connection(self) -> None:
         tensor = torch.zeros(10, 10)
         world_size = run_function(debug, tensor, dp=True)
 
         self.assertTrue(
             world_size >= 2, f"Need at least 2 gpus but was provided only {world_size}."
         )
 
-    def test_gradient_noclip_zeronoise(self):
+    def test_gradient_noclip_zeronoise(self) -> None:
         # Tests that gradient is the same with DP or with DDP
         weight_dp, weight_nodp = torch.zeros(10, 10), torch.zeros(10, 10)
 
@@ -272,7 +272,7 @@ def test_gradient_noclip_zeronoise(self):
 
         self.assertTrue(torch.norm(weight_dp - weight_nodp) < 1e-7)
 
-    def test_ddp_hook(self):
+    def test_ddp_hook(self) -> None:
         # Tests that the DDP hook does the same thing as naive aggregation with per layer clipping
         weight_ddp_naive, weight_ddp_hook = torch.zeros(10, 10), torch.zeros(10, 10)
 
@@ -297,7 +297,7 @@ def test_ddp_hook(self):
             f"DDP naive: {weight_ddp_naive}\nDDP hook: {weight_ddp_hook}",
         )
 
-    def test_add_remove_ddp_hooks(self):
+    def test_add_remove_ddp_hooks(self) -> None:
         remaining_hooks = {
             "attached": None,
             "detached": None,

opacus/tests/distributed_poisson_test.py

@@ -54,13 +54,13 @@ def setUp(self) -> None:
 
         self.samplers, self.dataloaders = self._init_data(seed=7)
 
-    def test_length(self):
+    def test_length(self) -> None:
         for sampler in self.samplers:
             self.assertEqual(len(sampler), 10)
         for dataloader in self.dataloaders:
             self.assertEqual(len(dataloader), 10)
 
-    def test_batch_sizes(self):
+    def test_batch_sizes(self) -> None:
         for dataloader in self.dataloaders:
             batch_sizes = []
             for x, _y in dataloader:
@@ -71,7 +71,7 @@ def test_batch_sizes(self):
                 np.mean(batch_sizes), self.batch_size // self.world_size, delta=2
             )
 
-    def test_separate_batches(self):
+    def test_separate_batches(self) -> None:
         indices = {
             rank: [i.item() for batch in self.samplers[rank] for i in batch]
             for rank in range(self.world_size)

opacus/tests/dpdataloader_test.py

@@ -20,12 +20,12 @@
 
 
 class DPDataLoaderTest(unittest.TestCase):
-    def setUp(self):
+    def setUp(self) -> None:
         self.data_size = 10
         self.dimension = 7
         self.num_classes = 11
 
-    def test_collate_classes(self):
+    def test_collate_classes(self) -> None:
         x = torch.randn(self.data_size, self.dimension)
         y = torch.randint(low=0, high=self.num_classes, size=(self.data_size,))
 
@@ -36,7 +36,7 @@ def test_collate_classes(self):
         self.assertEqual(x_b.size(0), 0)
         self.assertEqual(y_b.size(0), 0)
 
-    def test_collate_tensor(self):
+    def test_collate_tensor(self) -> None:
         x = torch.randn(self.data_size, self.dimension)
 
         dataset = TensorDataset(x)
@@ -46,7 +46,7 @@ def test_collate_tensor(self):
 
         self.assertEqual(s.size(0), 0)
 
-    def test_drop_last_true(self):
+    def test_drop_last_true(self) -> None:
         x = torch.randn(self.data_size, self.dimension)
 
         dataset = TensorDataset(x)

opacus/tests/grad_sample_module_test.py

@@ -225,26 +225,26 @@ def forward(self, x: torch.Tensor):
         register_grad_sampler(SimpleLinear)(compute_linear_grad_sample)
         GradSampleModule(SimpleLinear(4, 2))
 
-    def test_custom_module_validation(self):
+    def test_custom_module_validation(self) -> None:
         with self.assertRaises(NotImplementedError):
             GradSampleModule(mobilenet_v3_small())
 
-    def test_submodule_access(self):
+    def test_submodule_access(self) -> None:
         _ = self.grad_sample_module.fc1
         _ = self.grad_sample_module.fc2
 
         with self.assertRaises(AttributeError):
             _ = self.grad_sample_module.fc3
 
-    def test_state_dict(self):
+    def test_state_dict(self) -> None:
         gs_state_dict = self.grad_sample_module.state_dict()
         og_state_dict = self.original_model.state_dict()
         # check wrapped module state dict
         for key in og_state_dict.keys():
             self.assertTrue(f"_module.{key}" in gs_state_dict)
             assert_close(og_state_dict[key], gs_state_dict[f"_module.{key}"])
 
-    def test_load_state_dict(self):
+    def test_load_state_dict(self) -> None:
         gs_state_dict = self.grad_sample_module.state_dict()
         new_gs = GradSampleModule(
             SampleConvNet(), batch_first=False, loss_reduction="mean"
@@ -261,11 +261,11 @@ def test_load_state_dict(self):
 class EWGradSampleModuleTest(GradSampleModuleTest):
     CLS = GradSampleModuleExpandedWeights
 
-    def test_remove_hooks(self):
+    def test_remove_hooks(self) -> None:
         pass
 
-    def test_enable_hooks(self):
+    def test_enable_hooks(self) -> None:
         pass
 
-    def test_disable_hooks(self):
+    def test_disable_hooks(self) -> None:
         pass