Today I decided to go into the labs and run the program. Mostly to see if it runs as expected (i.e. without crashing or placing pieces in the same places). To my surprise, it worked well. Not great, of course, but you could evidence the agent learning. It learned that placing I-pieces vertically in special states and next to climbs of 2 or more is good and later it seemed to learn O-pieces go best on flat surfaces.
The only problem was one evidenced in a previous run. When the field gets too full, the new piece for the next field isn’t found and just drops down. Not a huge problem but something that must be tackled nonetheless.
Seems I don’t have to finish the last test case for proper functionality. I will of course though…