removed commented out numpy tests

Author: kashif

tests/test_scheduler.py

@@ -173,34 +173,6 @@ def test_step_shape(self):
             self.assertEqual(output_0.shape, sample.shape)
             self.assertEqual(output_0.shape, output_1.shape)
 
-    # def test_pytorch_equal_numpy(self):
-    #     kwargs = dict(self.forward_default_kwargs)
-
-    #     num_inference_steps = kwargs.pop("num_inference_steps", None)
-
-    #     for scheduler_class in self.scheduler_classes:
-    #         sample_pt = self.dummy_sample
-    #         residual_pt = 0.1 * sample_pt
-
-    #         sample = sample_pt.numpy()
-    #         residual = 0.1 * sample
-
-    #         scheduler_config = self.get_scheduler_config()
-    #         scheduler = scheduler_class(tensor_format="np", **scheduler_config)
-
-    #         scheduler_pt = scheduler_class(tensor_format="pt", **scheduler_config)
-
-    #         if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
-    #             scheduler.set_timesteps(num_inference_steps)
-    #             scheduler_pt.set_timesteps(num_inference_steps)
-    #         elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"):
-    #             kwargs["num_inference_steps"] = num_inference_steps
-
-    #         output = scheduler.step(residual, 1, sample, **kwargs).prev_sample
-    #         output_pt = scheduler_pt.step(residual_pt, 1, sample_pt, **kwargs).prev_sample
-
-    #         assert np.sum(np.abs(output - output_pt.numpy())) < 1e-4, "Scheduler outputs are not identical"
-
     def test_scheduler_outputs_equivalence(self):
         def set_nan_tensor_to_zero(t):
             t[t != t] = 0
@@ -304,10 +276,6 @@ def test_variance(self):
         assert torch.sum(torch.abs(scheduler._get_variance(487) - 0.00979)) < 1e-5
         assert torch.sum(torch.abs(scheduler._get_variance(999) - 0.02)) < 1e-5
 
-    # TODO Make DDPM Numpy compatible
-    def test_pytorch_equal_numpy(self):
-        pass
-
     def test_full_loop_no_noise(self):
         scheduler_class = self.scheduler_classes[0]
         scheduler_config = self.get_scheduler_config()
@@ -555,72 +523,6 @@ def full_loop(self, **config):
 
         return sample
 
-    # def test_pytorch_equal_numpy(self):
-    #     kwargs = dict(self.forward_default_kwargs)
-    #     num_inference_steps = kwargs.pop("num_inference_steps", None)
-
-    #     for scheduler_class in self.scheduler_classes:
-    #         sample_pt = self.dummy_sample
-    #         residual_pt = 0.1 * sample_pt
-    #         dummy_past_residuals_pt = [residual_pt + 0.2, residual_pt + 0.15, residual_pt + 0.1, residual_pt + 0.05]
-
-    #         sample = sample_pt.numpy()
-    #         residual = 0.1 * sample
-    #         dummy_past_residuals = [residual + 0.2, residual + 0.15, residual + 0.1, residual + 0.05]
-
-    #         scheduler_config = self.get_scheduler_config()
-    #         scheduler = scheduler_class(tensor_format="np", **scheduler_config)
-
-    #         scheduler_pt = scheduler_class(tensor_format="pt", **scheduler_config)
-
-    #         if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
-    #             scheduler.set_timesteps(num_inference_steps)
-    #             scheduler_pt.set_timesteps(num_inference_steps)
-    #         elif num_inference_steps is not None and not hasattr(scheduler, "set_timesteps"):
-    #             kwargs["num_inference_steps"] = num_inference_steps
-
-    #         # copy over dummy past residuals (must be done after set_timesteps)
-    #         scheduler.ets = dummy_past_residuals[:]
-    #         scheduler_pt.ets = dummy_past_residuals_pt[:]
-
-    #         output = scheduler.step_prk(residual, 1, sample, **kwargs).prev_sample
-    #         output_pt = scheduler_pt.step_prk(residual_pt, 1, sample_pt, **kwargs).prev_sample
-    #         assert np.sum(np.abs(output - output_pt.numpy())) < 1e-4, "Scheduler outputs are not identical"
-
-    #         output = scheduler.step_plms(residual, 1, sample, **kwargs).prev_sample
-    #         output_pt = scheduler_pt.step_plms(residual_pt, 1, sample_pt, **kwargs).prev_sample
-
-    #         assert np.sum(np.abs(output - output_pt.numpy())) < 1e-4, "Scheduler outputs are not identical"
-
-    # def test_set_format(self):
-    #     kwargs = dict(self.forward_default_kwargs)
-    #     num_inference_steps = kwargs.pop("num_inference_steps", None)
-
-    #     for scheduler_class in self.scheduler_classes:
-    #         scheduler_config = self.get_scheduler_config()
-    #         scheduler = scheduler_class(tensor_format="np", **scheduler_config)
-    #         scheduler_pt = scheduler_class(tensor_format="pt", **scheduler_config)
-
-    #         if num_inference_steps is not None and hasattr(scheduler, "set_timesteps"):
-    #             scheduler.set_timesteps(num_inference_steps)
-    #             scheduler_pt.set_timesteps(num_inference_steps)
-
-    #         for key, value in vars(scheduler).items():
-    #             # we only allow `ets` attr to be a list
-    #             assert not isinstance(value, list) or key in [
-    #                 "ets"
-    #             ], f"Scheduler is not correctly set to np format, the attribute {key} is {type(value)}"
-
-    #         # check if `scheduler.set_format` does convert correctly attrs to pt format
-    #         for key, value in vars(scheduler_pt).items():
-    #             # we only allow `ets` attr to be a list
-    #             assert not isinstance(value, list) or key in [
-    #                 "ets"
-    #             ], f"Scheduler is not correctly set to pt format, the attribute {key} is {type(value)}"
-    #             assert not isinstance(
-    #                 value, np.ndarray
-    #             ), f"Scheduler is not correctly set to pt format, the attribute {key} is {type(value)}"
-
     def test_step_shape(self):
         kwargs = dict(self.forward_default_kwargs)
 
@@ -953,28 +855,6 @@ def test_time_indices(self):
         for t in [0, 500, 800]:
             self.check_over_forward(time_step=t)
 
-    # def test_pytorch_equal_numpy(self):
-    #     for scheduler_class in self.scheduler_classes:
-    #         sample_pt = self.dummy_sample
-    #         residual_pt = 0.1 * sample_pt
-
-    #         sample = sample_pt.numpy()
-    #         residual = 0.1 * sample
-
-    #         scheduler_config = self.get_scheduler_config()
-    #         scheduler_config["tensor_format"] = "np"
-    #         scheduler = scheduler_class(**scheduler_config)
-
-    #         scheduler_config["tensor_format"] = "pt"
-    #         scheduler_pt = scheduler_class(**scheduler_config)
-
-    #         scheduler.set_timesteps(self.num_inference_steps)
-    #         scheduler_pt.set_timesteps(self.num_inference_steps)
-
-    #         output = scheduler.step(residual, 1, sample).prev_sample
-    #         output_pt = scheduler_pt.step(residual_pt, 1, sample_pt).prev_sample
-    #         assert np.sum(np.abs(output - output_pt.numpy())) < 1e-4, "Scheduler outputs are not identical"
-
     def test_full_loop_no_noise(self):
         scheduler_class = self.scheduler_classes[0]
         scheduler_config = self.get_scheduler_config()