Debugging a Model in a Supervised Setting

The first thing i check is, like, is the normalization of th different things correct? Andyet? If it's not, it's going to be very hard for the modelto learn. So that definitely took a tiny bit of debugging ening to ag another thing that turned out to be very important was having a reconstruction which encourages the agent to get all the information in its imputs through memory to its output. And some ofthe setting, fortunately, i was building on some prior work that had already shown that, and so i knew that that was something i should try. But without it, it didn't work.