PR Summary

This PR implements multiple improvements across the project:

1. A new notebook (geval_deepeval_integration_demo.ipynb) has been added to showcase the integration between DeepEval's GEval metric and the ValidMind library. The notebook covers installation instructions, setting up the environment, creating test cases, assigning scores, and even plotting the results with customized box plots.

2. Several dependency version updates have been applied in the poetry.lock and pyproject.toml files. Notable changes include downgrading or constraining versions of packages such as numpy, blis, datasets, fsspec, pyarrow, and related extras. This helps to ensure compatibility with supported Python versions and may resolve potential issues with newer releases.

3. The LLM dataset conversion function in validmind/datasets/llm/agent_dataset.py was refactored. The logic to convert test cases and goldens to a DataFrame has been divided into helper methods (_process_test_cases, _process_goldens) with a new helper (_add_optional_fields) to streamline processing optional fields. A safeguard has been added to return an empty row if no data is present.

4. In the plotting module (validmind/tests/plots/BoxPlot.py), minor modifications were made to reference the private dataframe attribute (_df) and adjust plot dimensions and spacing to improve visualization clarity.

5. A new scorer module for GEval (validmind/scorer/llm/deepeval/GEval.py) has been introduced, which integrates DeepEval’s functionality into the ValidMind testing framework. This scorer sets up evaluation using provided criteria, rubric, and evaluation steps, processes each test case from a dataset, and returns scores along with detailed explanations.

Test Suggestions

• Run the new GEval integration notebook to verify that test cases are scored correctly and that the resulting evaluations (score, reason, criteria) match expectations.
• Validate that the agent dataset conversion handles both complete and missing optional fields, ensuring that empty rows are generated if necessary.
• Execute the plotting tests (BoxPlot) to confirm that the updated plot dimensions and spacing yield clear and readable visual outputs.
• Perform dependency and compatibility tests to verify that the new version constraints (e.g., numpy, pyarrow, fsspec) work as intended in the supported Python versions.