- this tag is here so we can easily find custom losses for different types of problems
- When making your own custom loss, just make sure that it’s continuous
- so gradient descent can minimize it iteratively
- HOWEVER, some models (like random forests), they don’t need to be smooth
- since they evaluate the quality of a split (e.g. gini index)