Suppose we have a module for approximate memory-efficient statistical analysis which stores data in a set of logarithmic bins. However, around zero, depending on the data, it may be suitable to switch to linear interpolation (as in "no measurement is absolutely precise, why use so many bins").

For now, the proposed new() interface (has not been released to CPAN) is as follows:

• base - base of logarithmic intervals, i.e. upper_bound/lower_bound;
• precision - width of linear intervals around zero;
• zero_thresh - optional threshold at which to switch to linear, normally it's calculated dynamically (we want to switch where the linear and logarithmic intervals are of the same width).

I'm ok with my data model, but the parameter names seem a bit weird.

I would like to rename them to relative_precision, absolute_precision, and linear_threshold respectively. Does that look clear enough?

I was also thinking of absolute/relative error, but error is really variable and no more than half the precision. I think this could cause additional WTF.

Are there any better ideas?

The module in question is Statistics::Descriptive::LogScale. Here's the previous discussion.