We develop an AI for the Tetris Game. We fork the python Tetris from here: https://gist.github.com/silvasur/565419/7e044a90eb97eb67d600b2fb776000ba36f6fcc9. Our AI is inspired by this one: http://leeyiyuan.github.io/tetrisai/.
We develop our own AI and plug it into the python Tetris.
We trained our AI using Genetic Algorithm and the Cross Entropy method. All of it is written in the genetic.py file that will train population. The logs of our different training are available in the logs folder.
logs_v5.txt contains the logs of our lastest train using a limit of 500 pieces per game. We trained those population during about 4 days.
logs_v6.txt contains the logs of our lastest train using a limit of 1000 pieces per game. We trained it non-stop during 2 weeks...
In those two files, each line representing an individual is formated as following:
[score, [list of weights for each of the 34 heuristics]]
Ex:
[399000, [-2.100689734519437, -6.296337703002559, -3.6053570383018902, -5.174030015302046, -7.094388505722243, -7.163019117701906, -4.155623378939865, -7.470271464077066, -6.837060809015177, 4.055711464398931, -23.189118507055124, -17.22774140164603, -6.636151504678342, 3.169793808221626, -18.464118325281074, -3.892389642098916, -6.496728529483608, -9.168122545328442, -3.8031204486650156, -1.1073865709954989, -13.649143804127453, -4.615274667006191, 1.1085085503624723, -3.115040621666638, -6.324824019458616, -6.621636476173672, -3.356875697348909, -4.568217955358195, -7.329346044325321, -1.8518026355838344, 9.84688820557799, -8.194108483773855, -9.07005038806947, 3.149233388252534]]
The score is computed using the following formula :
nb_line_completed*1000 + nb_piece_dropped