Improve RankD Tests

[bbengfort]

Right now the Rank1D and Rank2D tests are very basic and can be improved using the new assert image similarity and pytest testing framework mechanisms.

Proposal/Issue

The test matrix should against the following:

• replace make_regression dataset with load_energy and update those tests
• use load_occupancy for classification tests
• algorithms: pearson, covariance, spearman (2D) and shaprio (1D)
• 1D case - horizontal orientation
• 1D case - vertical orientation
• Test that an exception is raised for unrecognized algorithms
• test the underlying rank matrix is correct

Unfortunately, we can't use pytest.mark.parametrize with visual test cases (yet), so we'll have to make individual tests for each.

Code Snippet

Tests will look approximately like:

def test_rank2d_bad_algorithm(self):
    """"
    Assert that unknown algorithms raise exception
    """"
    with pytest.raises(YellowbrickValueError, match="unknown algorithm"):
         # do the thing 

def test_rank2d_pearson_regression(self):
    """"
    Test Rank2D images similar with pearson scores on regression dataset 
    """"
    data = load_energy(return_dataset=True)
    oz = Rank2D(algorithm='pearson')
    oz.fit_transform(data) 
    npt.assert_array_equal(oz.ranks_, [[]])
    self.assert_images_similar(oz, tol=0.25) 

Background

See #68 and #429

[tabishsada]

I'll work on this.

[yanigisawa]

Are there additional tests / algorithms expected here beyond the Kendal-Tau metric added above?

[lwgray]

@yanigisawa Thanks for the comment. Did you have other algorithms in mind? Note that this issue is a little tricky as it's closely tied with #682 (updating tests to use the new datasets module) and #318 (using the new pytest testing style and fixtures). If you'd like to work on this issue, the rankd tests could certainly use beefing up, but make sure to take a look at those other two issues as well!