Three tasks for rebuttal

[vxbrandon]

Complete three tasks assigned by Mo here

• Add a boolean flag to turn on/off vectorization: Now by setting the vectorization flag to be true/false, you can turn on and off vectorization of impurity calculations and histogram insertions (classification only as we vectorized regression case before submission)
• Change the y axis of scaling experiments from wall-clock time to the number of insertions: After this change, we can observe that our algorithm is scaling with a logarithm of data size. Images below are the graphs I plot by running investigate_scaling.py and make_scaling_plot.py on my laptop.

• Turn off the divide-by-zero warning: turn off in the first line of the get_impurity_reductions function.
• Move all the log files to \logs directory.

[motiwari]

Overall ok but needs some fixes:

• With vectorization off, do we now get similar numbers to those presented in the original paper?

• Add a global USE_VECTORIZATION = False in constants.py and use it everywhere instead of default inline values throughout