Dataframes of EvalResult are not the same after (de)serializing

[tstadel]

Describe the bug
After calling EvaluationResult.save() and EvaluationResult.load() the underlying dataframes are not equal with respect to pd.testing.assert_frame_equal.
Specifically, they differ:

• None values get NaN values
• df.index changes from weird non-unique keys to a clean RangeIndex

Error message
If you rely on the same data/behavior after serialization, especially the first point causes troubles when you try to write the dataframe to a database (e.g. NaN values are not supported by SQLAlchemy)

Expected behavior
The underlying dataframes of EvaluationResult pass pd.testing.assert_frame_equal after serialization.

To Reproduce
Run test test_generative_qa_w_promptnode_eval and compare eval_result with saved_eval_result

FAQ Check

• Have you had a look at our new FAQ page?