grafx - Extensible Java Graphics Package Code

Adding computer-player-sample.mp4.

Tweaking the computer AI. These changes were done last night but I could not
commit. With these settings the computer player performs extremely well on
level 23; it seems to pass that level almost every time, with the following
block distribution:
1,2,3-er blocks: weight = 7
4,5-er blocks: weight = 5 or 10, depending on unique mirror image
6-er blocks: weight = 4 or 8, depending on unique mirror image
7-er block cube: weight = 19
The pit is 7x6x10, with 10 layers to clear. Please note that there are 165
unique 6-er blocks, there are 37 total 4+5-er blocks, and there are only four
1+2+3-er blocks - in other words almost EVERY SINGLE BLOCK in the distribution
is a 6-er block.
Also, "plug the hole" metric was changed to start from wall score being 7 or
more (all the way up to 12) in other words the effective score now is zero
through six, with 7 (e.g. 2,2,2,1) corresponding to 1, the lowest, and 12 (e.g.
3,3,3,3) corresponding to 6, the highest. Other values that have been tweaked:
OVERALL_HEIGHT_METRIC: 0.65 -> 0.70
COVERED_HOLE_METRIC: 3.00 -> 3.10
NOOK_AND_CRANNY_METRIC: 1.30 -> 1.35
PLUG_THE_HOLE_METRIC: 0.20 -> 0.40

Adding a BLO file of the computer playing a very hard level.

Upping the COVERED_HOLE_METRIC weight a massive, massive amount, and improving
the PLUG_THE_HOLE metric a bit, so that it's sensitive to a hole which is 2
units deep. The algorithm seems to perform superbly now on level 23.

Making some very minor tweaks. Trying to pass the ultra-hard level 24 on a
regular basis. Still no cigar.

Adding commented out code which prints the covered hole metric.

Adding commented out code for custom BlockSource.

Level 5, changing depth of pit from 7 to 9.

Whoa. This is difficult. The Eval function works well but there are some
loose ends to tie up. Documentation etc.

Adding launch-easy-level-test.sh.