One thing that trips any newcomer to DL and certainly deep reinforcement learning is when to know whether your model is failing, is just being a bit stubborn, or is not ever going to work. It is a question that causes frustration and angst in the AI field and often leaves you to wonder: what if I let that agent train a day longer? Unfortunately, if you speak to experts, they will often say just be patient and keep training, but this perhaps builds on those frustrations. After all, what if what you built has no hope of ever doing anything—are you wasting time and energy to keep it going?
Another issue that many face is that the more complex an algorithm/model gets, the more time it takes to train, except you never ...
