Potential bug in `evaluate_single_query`

[qlan3]

If I understand it correctly, for the preprocess in function evaluate_single_query, we need to convert response and answer to the same type before comparing.
However, after checking Line 140-148, I notice that answer_df is converted to str first:

WideSearch/src/evaluation/evaluation.py

Line 140 in 0e664c3

answer_df[col] = answer_df[col].astype(str)

But later it might be converted to float:

WideSearch/src/evaluation/evaluation.py

Line 147 in 0e664c3

answer_df[col] = answer_df[col].astype(float)

while response_df is always converted to str:

WideSearch/src/evaluation/evaluation.py

Line 148 in 0e664c3

response_df[col] = response_df[col].astype(str)

Is this a bug? If so, will it affect the final result greatly?

[jj-z-bd]

@qlan3
This section of logic isn't very standardized, but it won't cause bugs. The subsequent preprocessing process will convert the relevant columns back to numerical values or strings.
That said, there are some issues with the data conversion logic here. When using the exact_match method with the model output being a float, problems may occur, though rarely. We've coordinated and merged a PR to address this.