Grid search

[msluszniak]

Closes #148

I also added some minor improvements in linear and ridge regression

[josevalim]

Do we want to support version without options?

scoring_fun =
cond do
  is_function(scoring_fun, 3) ->
    fn x, y -> scoring_fun.(x, y, opts) end

  is_function(scoring_fun, 2) and opts != [] -> 
    raise ArgumentError, "no options must be given if scoring_fun expects two arguments"

  is_function(scoring_fun, 2) -> 
    scoring_fun

  true ->
    raise ArgumentError, "expected scoring_fun to be a function of arity 2 or 3, got: #{inspect(scoring_fun)}"
end

[josevalim]

Wait a second... we don't process the opts in anyway. So I don't see why we need to pass them here? If you need to pass options to the scoring_fun, you can just close over it:

     iex> scoring_fun = fn x, y ->
      ...>   {x_train, x_test} = x
      ...>   {y_train, y_test} = y
      ...>   model = Scholar.Linear.LinearRegression.fit(x_train, y_train, some_option: 123)

I think we can remove the opts argument in the cross validation functions now that we have weighted cross validate.

[josevalim]

I also posted an example on grid search. :)

[josevalim]

I have added some comments! Let's also add guards to those functions :)

[josevalim]

We need to update the docs and add simple tests for these changes. :)

[josevalim]

Also add guards for scoring funs here and below. And please don't forget the tests for the weights as tensors :D

[msluszniak]

Yes, I remember about them ;)

[msluszniak]

Actually, there were tests for weights as a tensor in linear regression. I added a test case where types of data and weights differ.