@tuannh99 while the code looks good I wonder what your use case is. As far as I get it you want to replace a float value with something else?
However, float values are notoriously difficult to compare and your are trying to convert a dense quantity into a categorical value which is ill-defined anyway without quantization.
To cut a long story short, please outline your use-case. If you really need this I would expect at least an option float_precision and maybe additionally float_rounding which can be floor, ceil and round at least...

@tuannh99 any news on this?

@srebhan I should have a commit to update this in a few.

@tuannh99 can you please comment on my concern above that trying to map a raw floating point value is very fragile and explain what your use-case is

This is fine as is. In a future PR we may want to expose the FormatFloat() precision argument, to give users the ability to control how precisely the floats are matched.