Setting up pytest

.github/workflows/run-pytest.yml

@@ -0,0 +1,23 @@
+name: Pytest
+on:
+  push:
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  run-pytest:
+    name: Execute Pytest tests
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Setup PDM
+        uses: pdm-project/setup-pdm@v4
+
+      - name: Install dependencies
+        run: make install-tools
+
+      - name: Run linters
+        run: make test

Makefile

@@ -23,6 +23,8 @@ install-tools: ## Install required utilities/tools
 	pdm run black --version
 	# check that correct Ruff version is installed
 	pdm run ruff --version
+	# check that Pytest is installed
+	pdm run pytest --version
 
 pdm-lock-check: ## Check that the pdm.lock file is in a good shape
 	pdm lock --check
@@ -47,7 +49,7 @@ format: install-deps-test ## Format the code into unified format
 verify:	install-deps-test ## Verify the code using various linters
 	pdm run black . --check
 	pdm run ruff check . --per-file-ignores=tests/*:S101 --per-file-ignores=scripts/*:S101
-	pdm run pylint src
+	pdm run pylint src tests
 
 requirements.txt:	pyproject.toml pdm.lock ## Generate requirements.txt file containing hashes for all non-devel packages
 	pdm export --prod --format requirements --output requirements.txt
@@ -58,6 +60,9 @@ verify-packages-completeness:	requirements.txt ## Verify that requirements.txt f
 distribution-archives: ## Generate distribution archives to be uploaded into Python registry
 	pdm run python -m build
 
+test: install-deps-test ## Execute tests with Pytest
+	pdm run pytest tests
+
 help: ## Show this help screen
 	@echo 'Usage: make <OPTIONS> ... <TARGETS>'
 	@echo ''

pyproject.toml

@@ -36,6 +36,7 @@ dev = [
     "mypy>=1.15.0",
     "ruff>=0.8.0",
     "pylint>=3.3.2",
+    "pytest>=8.3.2",
 ]
 
 [project.scripts]

tests/mocks/__init__.py

@@ -0,0 +1 @@
+"""Mocks for tests."""

tests/mocks/mock_classes.py

@@ -0,0 +1,18 @@
+"""Mocks for various classes for use during testing."""
+
+from dataclasses import dataclass
+from langchain_core.runnables import Runnable
+
+
+@dataclass
+class MockResult:
+    """Mock of LLM response"""
+
+    content: str
+
+
+class MockLLM(Runnable):
+    """Mock LLM class"""
+
+    def invoke(self, input, config=None, **kwargs):  # pylint: disable=W0622
+        return MockResult("5")

tests/utils/__init__.py

@@ -0,0 +1 @@
+"""Tests for evalutation utilities."""

tests/utils/test_similarity_score_llm.py

@@ -0,0 +1,12 @@
+"""Tests for similarity score utilities"""
+
+from road_core_eval.utils.similarity_score_llm import AnswerSimilarityScore
+from tests.mocks.mock_classes import MockLLM
+
+
+def test_similarity_score():
+    """Test similarity score calculation."""
+    mock_llm = MockLLM()
+
+    similarity_scorer = AnswerSimilarityScore(mock_llm)
+    assert similarity_scorer.get_score("question", "answer", "response") == 0.5