Home / One-Frame Breakout / 10x longer

MinAtar Breakout · follow-up

Breakout, 10x longer

Same game, players and protocol as One-Frame Breakout, but each life trains for 300 blocks (300,000 moments) instead of 30. 12 seeds per player. cadence-net 0.89.0.

The first run stopped while every curve was still climbing. So: what happens if you just keep going?

Bricks per test game

0 5 10 15 20 25 0 50 100 150 200 250 300 training block (1,000 moments each) first run ended here
Cadence, recall + position termCadence, recall, no position termCadence, no recall memoryPlain MLP, one framePlain MLP, given two frames

Median of 12 seeds, each point a 10-block window. Shaded: middle half of seeds for the position-term brain and the two-frame MLP.

What it showed

Numbers

player30 blocks300 blocks [IQR]agrees w/ coach
Cadence, recall + position term5.713.1 [11.8-13.7]94%
Cadence, recall, no position term5.610.4 [8.0-12.3]86%
Cadence, no recall memory7.013.3 [10.5-15.0]88%
Plain MLP, one frame5.16.0 [5.8-6.2]91%
Plain MLP, given two frames6.719.2 [17.7-20.4]90%

Bricks per test game over the last 10 blocks, median over seeds. The 30-block column is the first run (separate lives, same seeds).

paired, by seedwinsmedian diffWilcoxon p
Cadence, recall + position term vs Cadence, recall, no position term7/12+3.00.0884
Cadence, no recall memory vs Cadence, recall + position term6/12-0.61.0000
Cadence, recall + position term vs Plain MLP, one frame12/12+6.90.0005
Cadence, no recall memory vs Plain MLP, one frame12/12+7.20.0005
Plain MLP, given two frames vs Cadence, recall + position term12/12+6.20.0005
Plain MLP, given two frames vs Cadence, no recall memory11/12+5.90.0015

How it ran

One command on a desktop CPU, about 30 minutes per Cadence life and 3 per MLP life, 6 at a time: run.py --seeds 10-21 --arms pos32,pos0,norecall,mlp1,mlp2 --blocks 300 --procs 6 --out runs/long10x. Nothing else changed from the pre-declared protocol in run.py. Raw lives in runs/long10x.