I’m a mathematician and preparing for a new position in ML validation etc.
Since ML / DSci was not my primary area of research I’ve begun developing a library of useful approaches to common problems that will be useful in my new position — along with relevant examples, links to corresponding libraries, and methods in sklearn etc.
I’d be curious of your structure for dependencies in knowledge, I.e., how you handle the fact that a confusion matrix and decision boundaries are prerequisite to ROC. Also do you use a template for concepts/techniques etc?
I have a structure note corresponding to the problem/task type: “classification performance metrics”. This structure note links to notes each of which are approaches to solve the problem— i.e., I have a note linked to in this structure note called “classification performance via ROC”, and “classification performance via f-beta”
Currently I have three note types and corresponding templates:
- Problem / Task —a structure note which specifies the task and links to…
- Approaches / Solutions— a note which codified one approach to solving the problem in the associated structure/problem note. If the approach requires knowledge of ML concepts about which I am unfamiliar then I have
- Object / Concept note— a simple note with object definition, applications, example, quick facts, and relevant references.
I’m hoping that a core set of tasks/problems with solutions example’s and references will be helpful.
If you have any suggestions for alternative structures for this particular task-solution-reference use-case I’d be grateful to hear your thoughts.
Happy to share my templates if anyone is interested. Please do the same if you are working with a task/problem-approach/solution sort of structure.