[BE] migrate all assertRaises tests to OptimizerInfo test_errors (#116315)

janeyx99 · pytorchmergebot · commit 44b98c09ca90 · 2023-12-27T00:08:31.000Z
Removes a part of the sparse adam test and the following three tests: `test_fused_optimizer_raises`, `test_duplicate_params_across_param_groups`, `test_duplicate_params_in_one_param_group` ``` (pytorch-3.10) [janeyx@devgpu023.odn1 ~/local/pytorch (d2d129d)]$ python test/test_optim.py -k test_fused_optimizer_raises -k test_duplicate_params_across_param_groups -k test_duplicate_params_in_one_param_group /home/janeyx/.conda/envs/pytorch-3.10/lib/python3.10/site-packages/scipy/__init__.py:146: UserWarning: A NumPy version >=1.17.3 and <1.25.0 is required for this version of SciPy (detected version 1.26.0 warnings.warn(f"A NumPy version >={np_minversion} and <{np_maxversion}" ... ---------------------------------------------------------------------- Ran 3 tests in 0.023s OK ``` Increases coverage by testing the duplicate param tests on ALL the optims instead of just one each. Also fixes SparseAdam bug which was accidentally calling torch.unbind through list instead of putting params in a list. This bug was caught by migrating the weird warning stuff to just one easy warning context manager, which checks that nothing else gets raised. The new test_errors does not run slower than before, overhead is still king: ``` (pytorch-3.10) [janeyx@devgpu023.odn1 ~/local/pytorch (d2d129d)]$ python test/test_optim.py -k test_errors /home/janeyx/.conda/envs/pytorch-3.10/lib/python3.10/site-packages/scipy/__init__.py:146: UserWarning: A NumPy version >=1.17.3 and <1.25.0 is required for this version of SciPy (detected version 1.26.0 warnings.warn(f"A NumPy version >={np_minversion} and <{np_maxversion}" .......................... ---------------------------------------------------------------------- Ran 26 tests in 10.337s OK ``` Compared to test_errors BEFORE my commit :p ``` (pytorch-3.10) [janeyx@devgpu023.odn1 ~/local/pytorch (b47aa69)]$ python test/test_optim.py -k test_errors /home/janeyx/.conda/envs/pytorch-3.10/lib/python3.10/site-packages/scipy/__init__.py:146: UserWarning: A NumPy version >=1.17.3 and <1.25.0 is required for this version of SciPy (detected version 1.26.0 warnings.warn(f"A NumPy version >={np_minversion} and <{np_maxversion}" .............sssssssssssss ---------------------------------------------------------------------- Ran 26 tests in 11.980s OK (skipped=13) (pytorch-3.10) [janeyx@devgpu023.odn1 ~/local/pytorch (b47aa69)]$ ``` Pull Request resolved: #116315 Approved by: https://github.com/mikaylagawarecki
diff --git a/test/optim/test_optim.py b/test/optim/test_optim.py
@@ -906,14 +906,6 @@ def test_sparse_adam(self):
             sparse_only=True,
             maximize=True,
         )
-        import warnings
-        with warnings.catch_warnings(record=True) as ws:
-            SparseAdam(torch.zeros(3))
-            self.assertEqual(len(ws), 1)
-            for warning in ws:
-                self.assertEqual(len(warning.message.args), 1)
-                self.assertRegex(warning.message.args[0],
-                                 "Passing in a raw Tensor as ``params`` to SparseAdam ")
 
     # ROCm precision is too low to pass this test
     def test_adadelta(self):
@@ -1438,20 +1430,6 @@ def closure():
         self.assertEqual(type(res1), type(res2))
 
 
-    def test_duplicate_params_in_one_param_group(self):
-        param = Parameter(torch.randn(1))
-        with self.assertWarnsOnceRegex(UserWarning, '.*a parameter group with duplicate parameters.*'):
-            Adamax([param, param], lr=0.01)
-
-    def test_duplicate_params_across_param_groups(self):
-        param = Parameter(torch.randn(1))
-        self.assertRaisesRegex(
-            ValueError,
-            'some parameters appear in more than one parameter group',
-            lambda: Adadelta([{'params': param}, {'params': param}])
-        )
-
-
     def test_fused_optimizer_does_not_step_if_foundinf(self):
         if not torch.cuda.is_available():
             self.skipTest("CUDA is required.")
@@ -1621,14 +1599,6 @@ def local_post_hook(opt: Optimizer, args: Tuple[Any], kwargs: Dict[Any, Any]):
         opt2.step()
         self.assertListEqual(data, [0, 1, 2, 5, 0, 1, 2, 5, 0, 1, 2, 5])
 
-    def test_fused_optimizer_raises(self):
-        if not torch.cuda.is_available():
-            self.skipTest("Requires CUDA devices")
-        for optimizer_ctor in (Adam, AdamW):
-            with self.assertRaisesRegex(RuntimeError, "`fused` and `foreach` cannot be `True` together."):
-                optimizer_ctor([torch.empty((), device="cuda")], foreach=True, fused=True)
-            with self.assertRaisesRegex(RuntimeError, "`fused` does not support `differentiable`"):
-                optimizer_ctor([torch.empty((), device="cuda")], differentiable=True, fused=True)
 
     @staticmethod
     def _state_dict_pre_hook(optimizer: Optimizer) -> None:
diff --git a/test/test_optim.py b/test/test_optim.py
@@ -26,7 +26,6 @@ def test_optim_infos_do_not_specify_global_cliquey_kwargs(self, device, dtype, o
             self.assertFalse(any(f for f in global_cliquey_flags if f in optim_input.kwargs))
 
 
-    @onlyCPU
     @optims([optim for optim in optim_db if optim.optim_error_inputs_func is not None])
     def test_errors(self, device, dtype, optim_info):
         optim_cls = optim_info.optim_cls
@@ -36,12 +35,20 @@ def test_errors(self, device, dtype, optim_info):
             optim_input = error_input.optimizer_error_input
             params, kwargs = optim_input.params, optim_input.kwargs
             if error_input.error_on == OptimizerErrorEnum.CONSTRUCTION_ERROR:
-                with self.assertRaisesRegex(error_input.error_type, error_input.error_regex):
-                    optim_cls(params, **kwargs)
+                if issubclass(error_input.error_type, Warning):
+                    with self.assertWarnsRegex(error_input.error_type, error_input.error_regex):
+                        optim_cls(params, **kwargs)
+                else:
+                    with self.assertRaisesRegex(error_input.error_type, error_input.error_regex):
+                        optim_cls(params, **kwargs)
             elif error_input.error_on == OptimizerErrorEnum.STEP_ERROR:
                 optim = optim_cls(params, **kwargs)
-                with self.assertRaisesRegex(error_input.error_type, error_input.error_regex):
-                    optim.step()
+                if issubclass(error_input.error_type, Warning):
+                    with self.assertWarnsRegex(error_input.error_type, error_input.error_regex):
+                        optim.step()
+                else:
+                    with self.assertRaisesRegex(error_input.error_type, error_input.error_regex):
+                        optim.step()
             else:
                 raise NotImplementedError(f"Unknown error type {error_input.error_on}")
 
diff --git a/torch/optim/optimizer.py b/torch/optim/optimizer.py
@@ -260,6 +260,7 @@ def __init__(self, params: ParamsT, defaults: Dict[str, Any]) -> None:
                                "is deprecated. In the future, this will raise an error. "
                                "Please wrap your Tensor in an iterable instead."),
                               FutureWarning)
+                params = [params]
             else:
                 raise TypeError("params argument given to the optimizer should be "
                                 "an iterable of Tensors or dicts, but got " +
diff --git a/torch/testing/_internal/common_optimizers.py b/torch/testing/_internal/common_optimizers.py
