I plan to submit a new metric to Hugging Face that will compute
accuracy treating equivalent answers (as defined by is_equiv) as
correct. Without this change, Hugging Face users would have to
download the math directory and add it to their PYTHONPATH before
using the new metric. Instead, they'll be able to install the package
from github using pip.
As part of this change, I've also:
Added a line to README with install instructions
Renamed equivalent.py to math_equivalence.py
(math_equivalence is better as a package name than equivalent)
Prefixed the name of private functions with an underscore in math_equivalence.py
I plan to submit a new metric to Hugging Face that will compute accuracy treating equivalent answers (as defined by is_equiv) as correct. Without this change, Hugging Face users would have to download the math directory and add it to their PYTHONPATH before using the new metric. Instead, they'll be able to install the package from github using pip.
As part of this change, I've also: