Add missing model evaluation metrics

[mailletf]

Blog post about MLDB Hacktoberfest 2016

---

Rationale: Many Kaggle competitions use evaluation metrics that are currently not reported by MLDB. This makes it harder to use to solve competitions.

DOD: Add the following metrics to classifier evaluation:

• Regression
  • Mean absolute error: https://www.kaggle.com/wiki/MeanAbsoluteError
• Binary classification
  • Log loss: https://www.kaggle.com/wiki/LogarithmicLoss
• Multi-class classification
  • Log loss: https://www.kaggle.com/wiki/MultiClassLogLoss
  • Hamming Loss: https://www.kaggle.com/wiki/HammingLoss

They should be added in: https://github.com/mldbai/mldb/blob/master/plugins/accuracy.cc

Internal ticket: MLDB-1968

[jpilaul]

I can work on this starting with Binary Classification Log Loss