How to choose a Metric?

So after our study group session, @willsa14 and I were wondering if anyone has a cheat sheet that can assist with deciding which metric should be used within a regression model or a classification model.

For Example.

Regression Matrices Classification Matrices
Mean Square Error Binary Accuracy
Mean Absolute Error Categorical Accuracy
Mean Absolute Percentage Error Sparse Categorical Accuracy
Cosine Proximity Top K Categorical Accuracy
MSLE Accuracy (common use)
Log Cosh Error (Log Cosh) Sparse Top K Categorical Accuracy

Any good practices for determining when to use one over the other within a regression and classification matrice?