I have been using Deep Learning with Tensorflow for a pet project only for educational purpose, and have been able to obtain quite good results with a CNN of my own.
However, the process I used to get there is not satisfying, because it looks like trial/error based on basic metrics: performance on training set and evaluation sets. I do not know if this is efficient, or if some layers are useless or over/under sized. Same for the kernel sizes.
Do you have any book or article or blog to recommend on the subject, for learning to be a better practitioner ? I have read "The Bengio" ("Deep Learning") and another book in French about practicing Tensorflow ("Comprendre le Deep Learning: Une introduction aux réseaux de neurones"), but none gave a practical method. This looks like craftmanship only.