"EmpiricalDistribution" has various shortcomings

Class uses file IO (CM should not be concerned with data persistence)
Class uses the "java.net" API (ditto)

I'd think that the core functionality could be provided without supporting data loading (from a file or a URL).

Data input would be via any of

• Collection
• Iterable
• Stream (Java 8)

A redesign goal should be to make the class immutable.
E.g. data source must be set in the constructor (rather than via a "load" method); there would thus be a one-to-one correspondence between data source and distribution instance.