synthetic data TODOs

[mmp2]

1. We need a "switch" so that somebody running the code can only generate some of the examples and not all of them.
3. It will be good to also save the "original data" before noise and new dimensions are added. Why? the ML algorithms are supposed to remove the noise and the extra dimensions, and for simple examples we can compare the output of the algorithms with the clean data.
5. write out a descriptor file [optionally]. that would be a parameter file (see my email) containing all the params that were used generating the current data file. [LOWer PRIORITY]