Simplify restarts and checkpointing

[Balaji Pokuri]

Currently, it saves the state of the optimization through a dictionary. However, it can be more useful if it just saves the data (x,y) . During restart, it can use add_point() to add to the total population. This can enable:

• easy testing of hyperparameters
• reduction of runs, the data can be simple appended and not over-written. calculation of covariance matrix should be done after every restart.

[Balaji Pokuri]

It is better if we store the actual query point, rather than the aliased query point. This feature may need implementation and usage of alias_parameters() and unalias_parameters() so that the storage will be in actual dimensions and not in the parametric space we are optimizing.

[Balaji Pokuri]

Partially resolved in eb18954. Aliasing still underway

[Baskar Ganapathysubramanian]

Low priority till Dec 2017

[Balaji Pokuri]

Make it into def add_points_from_file()

[Balaji Pokuri]

current strategy is good enough, considering that we are not going to do alias-unalias method