OrdinalEncoder imputes missing values with 0 instead of -1

[janmotl]

Proposed fix:

1. Change a line with transformed_column.fillna(0, inplace=True).
2. Write a test to make sure that the error does not repeat.
3. Truncate test_inverse_transform() - it is ok to raise an exception when the data were imputed.

[janmotl]

However, that breaks other encoders:

    def test_backward_encoder(self):
        X = pd.DataFrame({'name': ['Afghanistan', 'Albania', 'Algeria'],
                          'landmass': [5, 4, 3]})
        X_t = pd.DataFrame({'name': ['American-Samoa'],
                            'landmass': [6]})

        enc = encoders.BackwardDifferenceEncoder()
        result = enc.fit(X).transform(X_t)

[janmotl]

Solved by @JohnnyC08 .