MachineJS Versions Save

Builds in support for Regression problems. Takes in an additional flag --join for performing sql-like joins on data. Switches to using scipy-sparse matrices (way better memory management).

As of this 0.9 release, we can:

1. Automatically read and format training data from a .csv file
2. Train and tune a neural network
3. Train and tune a random forest
4. Make predictions against a predictions dataset for each of these tuned classifiers
5. Create an ensemble of both these classifiers by averaging their predictions together