Merge pull request #87 from SFI-Visual-Intelligence/christian/update-dataloader-recall

Christian/update dataloader recall + a bit of the structure

Author: c-salomonsen

CollaborativeCoding/load_metric.py

@@ -107,7 +107,7 @@ def resetmetric(self):
         y_pred = th.rand((5, class_size))
 
         metricwrapper = MetricWrapper(
-            metric,
+            *metrics,
             num_classes=class_size,
             macro_averaging=True if class_size % 2 == 0 else False,
         )

tests/test_wrappers.py

@@ -1,16 +1,12 @@
 from pathlib import Path
-from tempfile import TemporaryDirectory
 
 import pytest
-import torch
-from torchvision import transforms
+import torch as th
 
 from CollaborativeCoding import MetricWrapper, load_data, load_model
 
 
 def test_load_model():
-    import torch as th
-
     image_shape = (1, 16, 16)
     num_classes = 4
 
@@ -36,17 +32,14 @@ def test_load_model():
 
 
 def test_load_data():
-    from tempfile import TemporaryDirectory
-
-    import torch as th
     from torchvision import transforms
 
     dataset_names = [
         "usps_0-6",
         "mnist_0-3",
         "usps_7-9",
         "svhn",
-        "mnist_4-9",  # Uncomment when implemented
+        "mnist_4-9",
     ]
 
     trans = transforms.Compose(
@@ -56,21 +49,16 @@ def test_load_data():
         ]
     )
 
-    with TemporaryDirectory() as tmppath:
-        for name in dataset_names:
-            dataset, _, _ = load_data(
-                name, train=False, data_dir=Path(tmppath), transform=trans
-            )
+    for name in dataset_names:
+        dataset = load_data(name, train=False, data_dir=Path.cwd() / "Data", transform=trans)
 
-            im, _ = dataset.__getitem__(0)
+        im, _ = dataset.__getitem__(0)
 
-            assert dataset.__len__() != 0
-            assert type(im) == th.Tensor and len(im.size()) == 3
+        assert dataset.__len__() != 0
+        assert type(im) is th.Tensor and len(im.size()) == 3
 
 
 def test_load_metric():
-    import torch as th
-
     metrics = ("entropy", "f1", "recall", "precision", "accuracy")
 
     class_sizes = [3, 6, 10]