In the "Metric" section on Page 5 of the paper, it is mentioned that:

"We categorize the synthesized equations into four buckets: Exact Solve, Almost Solve, Close, and Not Close. Exact Solve is quantitatively evaluated using a symbolic match. An equation is tagged as 'Almost Solve' if the dataset loss is small but the generated equation has an extra term or lacks one term. A Close equation captures the general structure of the solution (such as a square root nested in an exponential) but not more than that, and Not Close includes all equations that are far from the solution."

However, after reviewing the repository, I couldn't find the code for this metric. Could you kindly point me to the exact file or function where this logic is implemented?

Thank you!