Basic pip install code

[redhog]

• Generic code to train and save all models
• setup.py that runs the model train/save from the build command

[redhog]

@bitsofbits LogisticModel does not take the same arguments as LogisticScorer :( Why is this? Specifically, intercept.

[bitsofbits]

@egil, this is a sklearn thing. The sklearn way is to put parameters in the constructor, but all of the fitting happens in the fit method, so to make a fitted model one typicaly does mdl = Model(param, param).fit(X,y). So the Models are following the sklearn convention (which is much easier anyway since some of them are thin wrappers around sklearn models), but one needs to be able to set the intercept and coefs on the Scorers so I'm passing them in.

Do we need to harmonize the interfaces? I'm sure it's possible to hack coefs and intercept into the LogisticModel constructor somehow, but it might be kind of ugly.