Tolerances on (linear) equalities about validate HOT 8 CLOSED

Writing spurs thought :).

I guess within validate, for the purpose of confrontation we can do the former: translate to abs(x-y) < eps. Within validate.errloc we can do the second.

from validate.

Agreed! Some parts to rewrite equality expressions into inequalities are present in validate.errloc

from validate.

Another option could be to rewrite it into an all.equal statement which is arguably the most natural R thing to do

from validate.

I tried that :-) since it would work for nonnumerics as well, but all.equal expects two vectors of equal size. That would make translation rather awkward.

from validate.

I now added the option lin.eq.eps which, if non-zero, only causes linear rules to be translated to the abs-version. Since it is difficult to discover whether variables (or outcomes of R-statements) are numeric, and also since linear statements are very common I think this is a good compromise for now. Users can always do

some_nonlinear_complex_function_of(x) < their_own_epsilon

to formulate rules in the non-linear case., See upcoming push

from validate.

btw: the slack introduced in validate.errloc could be set to

max(lin.eq.eps, [the default for errloc])

but we'll have to see how user-friendly that is. I put the default for lin.eq.eps to 0 so users must explicitly set a tolerance. Not sure about that choice really. In the future we could possibly do this rule-wise as well.

from validate.

ok. Changing the default to 1e-8. Since this will not hurt comparisons between integer-rounded values, and works better in many cases of fractional numbers it is probably better. It is also reported in the summary of a confrontation object (since the actually evaluated expression is shown).

from validate.

for general equality rules we could also write our own eq function that can handle numeric and categorical data and takes a tolerance.

from validate.