trying to learn mlops (particulary data engineering). following madewithml's course. these are the following sections:
- old design (done). Justifying and planning developing a ML model. Figuring out requirements of model and system.
- old data (done) Pre-modeling workflow. Labeling, eda, preprocessing, augmentation and splitting.
- old modeling (done) Doing modeling better. Training (distributed and single gpu), simple baselines. Evaluating and debugging models. the course was updated, so I'll quickly run back through the updated sections
- design
- data
- model
- dev
- utilities
- testing
- reproducibility
- production also want to look into:
- read designing ml systems
- build and deploy something (what? I've deployed a couple models). should wrap in a useful UI- an app or website. maybe cancer detecting app (if its this, make sure they can download and run the model locally so those with limited internet access can get more value.) - more on cancer detector- it should find objects of interest, then return back to the user a % that is cancer (after applying some scaling technique) and some interpretability like lime.