Fix to Cosine Similiarity To Probability Clipping

[Unobtainiumrock]

What does this PR do?

This PR fixes an issue in the _convert_cosine_similarity_to_probability function where values outside the range of [-1, 1] were not clipped, leading to overflow problems during cosine similarity conversion in the retrieve_string_queries method.

Summary of Changes:

• Added np.clip to ensure that values in D are restricted to the range [-1, 1] before further calculations.
• This change addresses issues caused by FAISS Index returning -1 for indices that don't match during cosine similarity searches, ensuring compatibility with the expected value range for subsequent operations.

Before Fix:

The function did not clip D values, causing potential overflow issues when D = (D + 1) / 2 was performed.

After Fix:

The updated function now clips D to [-1, 1], preventing overflow errors and ensuring accurate conversion to probabilities:

def _convert_cosine_similarity_to_probability(self, D: np.ndarray) -> np.ndarray:
    D = np.clip(D, -1, 1)
    D = (D + 1) / 2
    D = np.round(D, 3)
    return D

Error Log Before Fix:

# (Include the relevant error log from the "Error Encountered" section of the image here)

Logs After Fix:

# (Include the log output from the "Log After Error Resolved" section of the image here)

Breaking Changes:

None. The changes are backward-compatible and resolve overflow issues without altering other functionality.

• Was this discussed/agreed via a GitHub issue?
  Yes.

• Did you read the contributor guideline?
  Yes.

• Did you make sure your PR does only one thing, instead of bundling different changes together?
  Yes.

• Did you make sure to update the documentation with your changes?
  Not applicable.

• Did you write any new necessary tests?
  Not applicable

• Did you verify new and existing tests pass locally with your changes?
  Yes.

• Did you list all the breaking changes introduced by this pull request?
  No breaking changes were introduced.

Additional Notes:

This fix ensures robustness and prevents errors during cosine similarity searches, especially when handling indices that do not match at all in FAISS Index. You can see more here https://docs.google.com/document/d/1ILCbNgrD6ILjHDHZV1rh7eKa-QPIHQYujoDj7q3nQ7E/edit?tab=t.0

Had fun solving this! 🙃

[liyin2015]

the code is fine, but need to set the formatting to be more consistent with the library.