Loss Functions and Metrics in Deep Learning