Especially if there are hundreds of data sets, I think we should rely on files
I don't think it's worth restricting the function to the linear tonemapper (it's even going to be almost impossible).
Regarding PBRT, I don't know your math knowledge, but it's a very valuable book which takes one step at a time trying to explain thinks in a pragmatic way so I find it a much better introduction than a paper like Veach's thesis which is another great reference but much more abstract.
Jeanphi
