Improvements to linear regression

[AtheMathmo]

The linear regression model is currently incomplete and untidy. The following changes are required:

• The LinRegressor field b is poorly named. Should be renamed to parameters. (Other suggestions welcome). This field should also be made private.
• Need to include intercept term. This means adding a new column of ones to the input data, and an additional parameter. For now we will do this by default.
• Need to implement the Optimization trait. Should also provide a method for optimized learning - train_with_optimization ?
• Within predict use an if let block instead of matching.
• Update the module description. Include example usage and update description to include added optimization techniques.

The linear regression model will see reworking in the future as we implement regularization (ridge regression) and model optimization (lasso, forward/back selection, etc.). These things are currently unmapped but the above changes will be needed regardless.

Please comment if you see other things which should be changed. LinRegressor was the first machine learning algorithm implemented in this library and so undoubtedly could use some work!

[AtheMathmo]

I handled the first change on the master branch as the others will take more time and thought but I wanted this one in soon.

[raulsi]

@AtheMathmo about second issue. new rows of ones in train function right. and an additional parameter is the default value will be 1 for now?

[AtheMathmo]

I mistyped, it should be a new column of ones. So if we have an input matrix X, we will produce the matrix [1 X] - where 1 represents a column of ones (with the same number of rows as X).

We'll need to do this change in both the train and predict functions. It will be sufficient to simply update X with something like:

let ones = Matrix::<f64>::ones(inputs.rows(), 1);
let full_inputs = ones.hcat(inputs);

And then use full_inputs in place of inputs for the rest of the scope.

I've added another improvement to the issue above as well (another easy one).

[AtheMathmo]

Closing this issue for now. I want to change how the optimization is handled in the future but for now this will lay the foundation.