Add basic test for classification

Author: akx

.gitignore

@@ -1,2 +1,3 @@
 *.py[cod]
 .*cache
+.valohai

README.md

@@ -14,3 +14,11 @@ Linting/formatting happens via `pre-commit`. Install it with `pip install pre-co
 
 The linters run by `pre-commit` are `ruff`, `black`, and `prettier`;
 you can (should) set up your IDE to run them automatically too.
+
+### Tests
+
+You can run tests with `py.test`:
+
+```
+py.test -v .
+```

conftest.py

@@ -0,0 +1,8 @@
+import pytest
+from valohai.internals import global_state
+
+
+@pytest.fixture()
+def valohai_utils_global_state():
+    global_state.flush_global_state()
+    return global_state

models/nlp/classification/huggingface/test_huggingface.py

@@ -0,0 +1,36 @@
+import csv
+
+import valohai
+
+# Snippet from https://valohai-ecosystem-datasets.s3.eu-west-1.amazonaws.com/yelp_reviews_batch_inference.txt
+EXAMPLE_DATA = """
+Old school.....traditional "mom 'n pop" quality and perfection.
+A great out of the way, non-corporate, vestige of Americana. You will love it.
+Good fish sandwich.
+I always feel like I am constantly bashing breweries for their food, but in my opinion, I feel the bar is raised for places like this.
+I called to complain, and the "manager" didn't even apologize!!! So frustrated. Never going back.  They seem overpriced, too.
+""".strip()
+
+
+def test_inference(valohai_utils_global_state, monkeypatch, tmp_path):
+    monkeypatch.setenv("VH_OUTPUTS_DIR", str(tmp_path))
+    input_path = tmp_path / "input.txt"
+    input_path.write_text(EXAMPLE_DATA)
+    valohai.prepare(
+        step="huggingface-classification-inference",
+        default_parameters={
+            "log_frequency": 1,
+            # This is an untrained model, so the results won't be very interesting.
+            "huggingface_repository": "distilbert-base-uncased",
+            "output_path": "test.csv",
+        },
+        default_inputs={
+            "data": str(input_path),
+        },
+    )
+    from models.nlp.classification.huggingface.inference import main
+
+    main()
+    with (tmp_path / "test.csv").open() as f:
+        results = list(csv.DictReader(f))
+    assert len(results) == 5

mypy.ini

@@ -11,3 +11,6 @@ ignore_missing_imports = True
 
 [mypy-datasets.*]
 ignore_missing_imports = True
+
+[mypy-pytest.*]
+ignore_missing_imports = True

pytest.ini

@@ -0,0 +1,4 @@
+[pytest]
+# this is needed because models.nlp has its own utils which would conflict
+# with the utils in the root directory with the default `prepend` mode
+addopts = --import-mode=append

ruff.toml

@@ -20,3 +20,6 @@ ignore = [
     "T2",
     "TRY003",
 ]
+
+[per-file-ignores]
+"**/test*.py" = ["S101"]  # tests can use assertions