Not sure whether this is a fitting place to ask such a question, but here goes…
I learned about simple machine learning and supervised/unsupervised training back in Uni, where I specialized in machine learning, but haven’t been able to utilize/practice that in my day-to-day work – which I am trying to do now.
What I seem to struggle with is understanding how one is supposed to work iteratively/logically when training a model and trying to apply it.
Today I tried to train an object detection model which couldn’t detect anything and had 0% accuracy after a day of labeling data and an hour of model training (using the ml.net wizard).
My current theory is that my training failed due to not having enough data. But how do I prove my thesis without (any data) having to spend another day labeling data, another hour (or more) training, to either prove or disprove my theory?
If a similar problem arose with a software solution, my approach would be to try to confirm my theory with a formal test/proof of concept (POC) to then confirm that what I observe is due to either A or B. It may take a while to set up a scenario to confirm/disprove my theory, but at least there’s a logical approach to things, and one is not bound to go with a solution 100%, only to realize it wasn’t the right decision.
How is the iterative process in machine learning?