After some modifications to the code, I took it in for another test run. Didn’t work, as kinda expected.
Here’s the low-down:
- Still having problems with tall states (culled states). FIXED! See findPieceRewards. It was using find RelevantStates and passing in a culled list of states. This has been fixed in findRelevantStates to accept culled and non-culled lists of SubStates.
- Also, pieces still seem to like certain positions. This is due to the list of bestPositions being of size 1, when it should be much bigger. FIXED – Sorta… It’s tested, and the test says it works fine, but I haven’t changed anything. Warrants further investigation.
- When an episode is over (the pieces spill over the top), the agent tries to find the new piece, but can’t and so the new piece falls down idly.
- Crashes on the 5th line of chaseGoal() with NullPointerException. FIXED! Was using i < iterates instead of i <= iterates.
- Height culling needs reworking due to states that have towers of width 3 or less being included still.
- findPieceRewards still not working as it seems to ignore height culling.
I should try to set up this environment at home so I don’t have to come to the lab every time I need to check stuff.