I think this is along the right lines, but I don't much like the +-1 parameter: especially as it's a floating point type argument. How about using an enum here:
enum extreme_value_type { type_1_maximum = 1, type_1_minimum = -1 };
I had actually considered doing that, but then started feeling a little ambiguous about it since I would have been using the enumerated value in a mathematical context. But, it does make a lot more sense.
All these need tests and sanity checks: preferably some sanity checks from an independent source if we can find one. Mathematica would be a good choice, but I don't have access to that here :-(
I don't think my department has an installation of Mathematica, but I do know a couple of math grads. I'll see what I can come up with.
Median looks like it needs adjusting to reflect the "sign". Skewness for min case is presumably the negative of the max case (needs checking)?
Yup... I overlooked a negative sign from the MathWorld site.
BTW the sandbox/math_toolkit directory is open for breaking changes again now (Boost.Math in 1.35 is effectively frozen bar trivial fixes), so when you have these worked up feel free to commit to the sandbox. Also let me know if you get stumped for test cases.
Can do. I'm mostly waiting on my student to finish off the functions for the skew normal. After that, I'll push those distributions up to the sandbox along with the random numbers generators. On a side note, I think it might be interesting to provide a generator () function for distributions that could return random variates. Andrew Sutton asutton@cs.kent.edu