simplify/speedup catalyst tests (#327)

tests/test_catalyst.py

@@ -1,11 +1,9 @@
 import os
 
+import catalyst
 import pytest
+import torch
 from catalyst import dl
-from catalyst.contrib.datasets import MNIST
-from catalyst.utils.torch import get_available_engine
-from torch import nn, optim
-from torch.utils.data import DataLoader
 
 from dvclive import Live
 from dvclive.catalyst import DvcLiveCallback
@@ -14,38 +12,46 @@
 # pylint: disable=redefined-outer-name, unused-argument
 
 
-@pytest.fixture(scope="session")
-def loaders(tmp_path_factory):
-    path = tmp_path_factory.mktemp("catalyst_mnist")
-    train_data = MNIST(path, train=True, download=True)
-    valid_data = MNIST(path, train=False, download=True)
-    return {
-        "train": DataLoader(train_data, batch_size=32),
-        "valid": DataLoader(valid_data, batch_size=32),
-    }
-
-
 @pytest.fixture
 def runner():
     return dl.SupervisedRunner(
-        engine=get_available_engine(),
+        engine=catalyst.utils.torch.get_available_engine(cpu=True),
         input_key="features",
         output_key="logits",
         target_key="targets",
         loss_key="loss",
     )
 
 
-def test_catalyst_callback(tmp_dir, runner, loaders):
-    model = nn.Sequential(nn.Flatten(), nn.Linear(28 * 28, 10))
-    criterion = nn.CrossEntropyLoss()
-    optimizer = optim.Adam(model.parameters(), lr=0.02)
+# see:
+# https://github.com/catalyst-team/catalyst/blob/e99f9/tests/catalyst/callbacks/test_batch_overfit.py
+@pytest.fixture
+def runner_params():
+    from torch.utils.data import DataLoader, TensorDataset
+
+    catalyst.utils.set_global_seed(42)
+    num_samples, num_features = int(32e1), int(1e1)
+    X, y = torch.rand(num_samples, num_features), torch.rand(num_samples)
+    dataset = TensorDataset(X, y)
+    loader = DataLoader(dataset, batch_size=32, num_workers=0)
+    loaders = {"train": loader, "valid": loader}
+
+    model = torch.nn.Linear(num_features, 1)
+    criterion = torch.nn.MSELoss()
+    optimizer = torch.optim.Adam(model.parameters())
+    scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, [3, 6])
+    return {
+        "model": model,
+        "criterion": criterion,
+        "optimizer": optimizer,
+        "scheduler": scheduler,
+        "loaders": loaders,
+    }
 
+
+def test_catalyst_callback(tmp_dir, runner, runner_params):
     runner.train(
-        model=model,
-        criterion=criterion,
-        optimizer=optimizer,
-        loaders=loaders,
+        **runner_params,
         num_epochs=2,
         callbacks=[
             dl.AccuracyCallback(input_key="logits", target_key="targets"),
@@ -70,17 +76,9 @@ def test_catalyst_callback(tmp_dir, runner, loaders):
     assert any("accuracy" in x.name for x in valid_path.iterdir())
 
 
-def test_catalyst_model_file(tmp_dir, runner, loaders):
-    model = nn.Sequential(nn.Flatten(), nn.Linear(28 * 28, 10))
-    criterion = nn.CrossEntropyLoss()
-    optimizer = optim.Adam(model.parameters(), lr=0.02)
-
+def test_catalyst_model_file(tmp_dir, runner, runner_params):
     runner.train(
-        model=model,
-        engine=runner.engine,
-        criterion=criterion,
-        optimizer=optimizer,
-        loaders=loaders,
+        **runner_params,
         num_epochs=2,
         callbacks=[
             dl.AccuracyCallback(input_key="logits", target_key="targets"),