adds checks for used device in metrics tests. #3335 (#3353)

Author: BanzaiTokyo

tests/ignite/metrics/gan/test_inception_score.py

@@ -32,6 +32,7 @@ def test_inception_score(available_device):
 
     p_yx = torch.rand(20, 3, 299, 299)
     m = InceptionScore(device=available_device)
+    assert m._device == torch.device(available_device)
     m.update(p_yx)
     assert isinstance(m.compute(), float)
 

tests/ignite/metrics/test_accuracy.py

@@ -66,6 +66,7 @@ def test_binary_wrong_inputs():
 @pytest.mark.parametrize("n_times", range(3))
 def test_binary_input(n_times, available_device, test_data_binary):
     acc = Accuracy(device=available_device)
+    assert acc._device == torch.device(available_device)
 
     y_pred, y, batch_size = test_data_binary
     acc.reset()
@@ -104,6 +105,7 @@ def test_multiclass_wrong_inputs():
 @pytest.mark.parametrize("n_times", range(3))
 def test_multiclass_input(n_times, available_device, test_data_multiclass):
     acc = Accuracy(device=available_device)
+    assert acc._device == torch.device(available_device)
 
     y_pred, y, batch_size = test_data_multiclass
     acc.reset()
@@ -155,6 +157,7 @@ def test_multilabel_wrong_inputs():
 @pytest.mark.parametrize("n_times", range(3))
 def test_multilabel_input(n_times, available_device, test_data_multilabel):
     acc = Accuracy(is_multilabel=True, device=available_device)
+    assert acc._device == torch.device(available_device)
 
     y_pred, y, batch_size = test_data_multilabel
     if batch_size > 1:

tests/ignite/metrics/test_average_precision.py

@@ -85,6 +85,7 @@ def test_data_binary_and_multilabel(request):
 def test_binary_and_multilabel_inputs(n_times, available_device, test_data_binary_and_multilabel):
     y_pred, y, batch_size = test_data_binary_and_multilabel
     ap = AveragePrecision(device=available_device)
+    assert ap._device == torch.device(available_device)
     ap.reset()
     if batch_size > 1:
         n_iters = y.shape[0] // batch_size + 1
@@ -129,6 +130,7 @@ def update_fn(engine, batch):
     engine = Engine(update_fn)
 
     ap_metric = AveragePrecision(device=available_device)
+    assert ap_metric._device == torch.device(available_device)
     ap_metric.attach(engine, "ap")
 
     np_y = y.numpy()

tests/ignite/metrics/test_cosine_similarity.py

@@ -45,6 +45,7 @@ def test_compute(n_times, test_case: Tuple[Tensor, Tensor, float, int], availabl
     y_pred, y, eps, batch_size = test_case
 
     cos = CosineSimilarity(eps=eps, device=available_device)
+    assert cos._device == torch.device(available_device)
 
     cos.reset()
     if batch_size > 1:
@@ -69,6 +70,7 @@ def test_compute(n_times, test_case: Tuple[Tensor, Tensor, float, int], availabl
 
 def test_accumulator_detached(available_device):
     cos = CosineSimilarity(device=available_device)
+    assert cos._device == torch.device(available_device)
 
     y_pred = torch.tensor([[2.0, 3.0], [-2.0, 1.0]], dtype=torch.float)
     y = torch.ones(2, 2, dtype=torch.float)

tests/ignite/metrics/test_precision.py

@@ -105,6 +105,7 @@ def ignite_average_to_scikit_average(average, data_type: str):
 @pytest.mark.parametrize("average", [None, False, "macro", "micro", "weighted"])
 def test_binary_input(n_times, available_device, average, test_data_binary):
     pr = Precision(average=average, device=available_device)
+    assert pr._device == torch.device(available_device)
     assert pr._updated is False
     y_pred, y, batch_size = test_data_binary
 
@@ -193,6 +194,7 @@ def test_multiclass_wrong_inputs():
 @pytest.mark.parametrize("average", [None, False, "macro", "micro", "weighted"])
 def test_multiclass_input(n_times, available_device, average, test_data_multiclass):
     pr = Precision(average=average, device=available_device)
+    assert pr._device == torch.device(available_device)
     assert pr._updated is False
 
     y_pred, y, batch_size = test_data_multiclass
@@ -260,6 +262,7 @@ def to_numpy_multilabel(y):
 @pytest.mark.parametrize("average", [None, False, "macro", "micro", "weighted", "samples"])
 def test_multilabel_input(n_times, available_device, average, test_data_multilabel):
     pr = Precision(average=average, is_multilabel=True, device=available_device)
+    assert pr._device == torch.device(available_device)
     assert pr._updated is False
 
     y_pred, y, batch_size = test_data_multilabel

tests/ignite/metrics/test_psnr.py

@@ -53,6 +53,7 @@ def test_psnr(test_data, available_device):
     data_range = (y.max() - y.min()).cpu().item()
 
     psnr = PSNR(data_range=data_range, device=available_device)
+    assert psnr._device == torch.device(available_device)
     psnr.update(test_data)
     psnr_compute = psnr.compute()
 

tests/ignite/metrics/test_recall.py

@@ -108,6 +108,7 @@ def ignite_average_to_scikit_average(average, data_type: str):
 @pytest.mark.parametrize("average", [None, False, "macro", "micro", "weighted"])
 def test_binary_input(n_times, available_device, average, test_data_binary):
     re = Recall(average=average, device=available_device)
+    assert re._device == torch.device(available_device)
     assert re._updated is False
 
     y_pred, y, batch_size = test_data_binary
@@ -195,6 +196,7 @@ def test_multiclass_wrong_inputs():
 @pytest.mark.parametrize("average", [None, False, "macro", "micro", "weighted"])
 def test_multiclass_input(n_times, available_device, average, test_data_multiclass):
     re = Recall(average=average, device=available_device)
+    assert re._device == torch.device(available_device)
     assert re._updated is False
 
     y_pred, y, batch_size = test_data_multiclass
@@ -263,6 +265,7 @@ def to_numpy_multilabel(y):
 def test_multilabel_input(n_times, available_device, average, test_data_multilabel):
 
     re = Recall(average=average, is_multilabel=True, device=available_device)
+    assert re._device == torch.device(available_device)
     assert re._updated is False
 
     y_pred, y, batch_size = test_data_multilabel

tests/ignite/metrics/test_roc_auc.py

@@ -87,6 +87,7 @@ def test_data_binary_and_multilabel(request):
 def test_binary_and_multilabel_inputs(n_times, available_device, test_data_binary_and_multilabel):
     y_pred, y, batch_size = test_data_binary_and_multilabel
     roc_auc = ROC_AUC(device=available_device)
+    assert roc_auc._device == torch.device(available_device)
     roc_auc.reset()
     if batch_size > 1:
         n_iters = y.shape[0] // batch_size + 1
@@ -147,6 +148,7 @@ def update_fn(engine, batch):
     engine = Engine(update_fn)
 
     roc_auc_metric = ROC_AUC(device=available_device)
+    assert roc_auc_metric._device == torch.device(available_device)
     roc_auc_metric.attach(engine, "roc_auc")
 
     np_y = y.numpy()

tests/ignite/metrics/test_ssim.py

@@ -163,7 +163,7 @@ def test_ssim_variable_batchsize(available_device):
     sigma = 1.5
     data_range = 1.0
     ssim = SSIM(data_range=data_range, sigma=sigma, device=available_device)
-
+    assert ssim._device == torch.device(available_device)
     y_preds = [
         torch.rand(12, 3, 28, 28, device=available_device),
         torch.rand(12, 3, 28, 28, device=available_device),
@@ -229,6 +229,7 @@ def test_ssim_uint8(available_device, shape, kernel_size, gaussian, use_sample_c
     sigma = 1.5
     data_range = 255
     ssim = SSIM(data_range=data_range, sigma=sigma, device=available_device)
+    assert ssim._device == torch.device(available_device)
     ssim.update((y_pred, y))
     ignite_ssim = ssim.compute()