[0.2.dev9] Binary logit & Regression Add ins

[cvanoli]

On this PR I'm proposing two small changes:

• In the binary logit run() function, the df is built with the get_data() function. To the parameters that are passed into this function, I'm adding to the columns to add in the df, the out_columns of the binary model, extra_columns = self.out_column, since this column is then used to set the out_true and out_false result values. I have encountered the issue with a household_relocation_model, that the building_id which is the out_column, was not appearing in the df, because it was not present in the model_expression nor the out_filters. This caused the update of the column full of null values.

• In the regression models, I've added the lines to save the residuals when the fitting is done, so the modeler can get the residuals with model.residuals for evaluation. The residuals are created using .resid which is an element of the RegressionModel object.

[coveralls]

Coverage increased (+0.1%) to 91.909% when pulling 061fabf on binary_logit_addin into 03d377e on master.

[smmaurer]

Hi @cvanoli, this looks great! Thanks for the bug fix and for the improvement to the diagnostic output. It seems like the tests were only failing because of a one-off glitch; I re-started the Travis job and it passed.

I'm going to add a commit updating the changelog, and then merge this!