Improvement concerning single sensitive attribute with multiple values

[eustomaqua]

I found two issues when I was using AdaFair:

1. ZeroDivisionError: division by zero may happen if the denominator is zero, such as

		tpr_protected = tp_protected / (tp_protected + fn_protected)
		tnr_protected = tn_protected / (tn_protected + fp_protected)

		tpr_non_protected = tp_non_protected / (tp_non_protected + fn_non_protected)
		tnr_non_protected = tn_non_protected / (tn_non_protected + fp_non_protected)

		alpha = 1 * (
				np.log((1. - estimator_error) / estimator_error) +
				np.log(n_classes - 1.))

			TPR = (float(tp)) / (tp + fn)
			TNR = (float(tn)) / (tn + fp)
			cumulative_balanced_error = 1 - (TPR + TNR) / 2
			cumulative_error = 1 - (float(tp) + float(tn)) / (tp + tn + fp + fn)

2. AdaFair only works for the situation where only one sensitive attribute is considered and that attribute has binary values (such as {0,1})

So I made two modifications for these two issues, alongside relevant lines using them

1. the check_zero function between line 42 and line 43
2. the _member_found function between line 331 and line 334
   Note that the _member_found function makes AdaFair suitable for situations where one sensitive attribute has multiple values (such as {0,1,2,3}), instead of only two optional values like {0,1}