Add tests to check how transformations are applied.

Author: arthurdouillard

tests/test_transforms.py

@@ -0,0 +1,91 @@
+import numpy as np
+import pytest
+from torchvision.transforms import ToTensor
+from torch.utils.data import DataLoader
+import torch
+
+from continuum.datasets import InMemoryDataset
+from continuum.scenarios import ClassIncremental, InstanceIncremental, ContinualScenario
+
+
+@pytest.fixture
+def dataset():
+    x = np.random.randint(0, 255, (100, 4, 4, 3), dtype=np.uint8)
+    y = np.random.randint(0, 3, (100,), dtype=np.int16)
+    t = np.ones_like(y)
+
+    t[:30] = 0
+    t[30:60] = 1
+    t[60:] = 2
+
+    return InMemoryDataset(x, y, t)
+
+
+
+@pytest.mark.parametrize("scenario,opt", [
+    (ClassIncremental, {'increment': 1}),
+    (InstanceIncremental, {}),
+    (ContinualScenario, {})
+])
+def test_same_transforms(dataset, scenario, opt):
+    trsfs = [
+        ToTensor(),
+        lambda tensor: tensor.fill_(0)
+    ]
+    s = scenario(dataset, transformations=trsfs, **opt)
+
+    for taskset in s:
+        loader = DataLoader(taskset)
+        for x, _, _ in loader:
+            assert torch.unique(x).numpy().tolist() == [0]
+
+
+@pytest.mark.parametrize("scenario,opt", [
+    (ClassIncremental, {'increment': 1}),
+    (InstanceIncremental, {}),
+    (ContinualScenario, {})
+])
+def test_diff_transforms(dataset, scenario, opt):
+    trsfs = [
+        [ToTensor(), lambda tensor1: tensor1.fill_(0)],
+        [ToTensor(), lambda tensor2: tensor2.fill_(1)],
+        [ToTensor(), lambda tensor3: tensor3.fill_(2)],
+    ]
+    s = scenario(dataset, transformations=trsfs, **opt)
+
+    for taskid, taskset in enumerate(s):
+        loader = DataLoader(taskset)
+        for x, _, _ in loader:
+            assert torch.unique(x).numpy().tolist() == [taskid]
+
+
+@pytest.mark.parametrize("scenario,opt,error", [
+    (ClassIncremental, {'increment': 1}, True),
+    (InstanceIncremental, {}, True),
+    (ContinualScenario, {}, True),
+    (ClassIncremental, {'increment': 1}, False),
+    (InstanceIncremental, {}, False),
+    (ContinualScenario, {}, False)
+])
+def test_diff_transforms_slice(dataset, scenario, opt, error):
+    trsfs = [
+        [ToTensor(), lambda tensor1: tensor1.fill_(0)],
+        [ToTensor(), lambda tensor2: tensor2.fill_(1)],
+        [ToTensor(), lambda tensor3: tensor3.fill_(2)],
+    ]
+    s = scenario(dataset, transformations=trsfs, **opt)
+
+    for taskid in range(len(s)):
+        if not error:
+            taskset = s[taskid]
+            loader = DataLoader(taskset)
+            for x, _, _ in loader:
+                assert torch.unique(x).numpy().tolist() == [taskid]
+        else:
+            with pytest.raises(ValueError):
+                s[:taskid]
+
+
+
+
+