Add a 3rd output class, labeled UNKNOWN.

[cbaakman]

When calculating MCC, accuracy and ROC AUC values in tensorboard, consider the following 3 class labels:

• BENIGN: when the CNN outputs a higher number on the left side, than on the right side
• PATHOGENIC: when the CNN outputs a higher number on the right side, than on the left side
• UNKNOWN: when the absolute difference between the CNN output numbers is below 0.5

BENIGN and PATHOGENIC both contribute to the MCC, accuracy and ROC values. Outputs of UNKNOWN however, are left out of the equation. They don't contribute to these scores. Furthermore, loss values are still calculated in the same way as before. So this change will not affect CNN training. Only the scores are calculated in a different way. @rgayatri is this what you intended?

[coveralls]

Pull Request Test Coverage Report for Build 3149893130

• 85 of 90 (94.44%) changed or added relevant lines in 5 files are covered.
• 6 unchanged lines in 1 file lost coverage.
• Overall coverage decreased (-0.01%) to 75.75%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
deeprank/models/metrics.py	45	47	95.74%
test/tools/test_metrics_tools.py	8	11	72.73%

Files with Coverage Reduction	New Missed Lines	%
deeprank/models/metrics.py	6	90.18%

Totals
Change from base Build 2971424162:	-0.01%
Covered Lines:	3030
Relevant Lines:	4000

---

💛 - Coveralls