When developing ML/DL models, I often have to make decisions about architecture (layers, loss function, shape), preprocessing method (tiling vs. resizing, text embeddings algorithm), optimization algorithm (SGD or ADAM) and hyperparameters choice. In many library/research papers, this decision seems to be made at random. Do we have collective sources of things people have tried that work? Where do you discuss best practices and techniques that sometimes make a lot of difference in developing a good model?
No comments yet.