Skip to content

zhoubin-me/mxdqn

 
 

Repository files navigation

mxdqn

MXNet implementation of Deep Q-learning

Result

Game Random Play Best Linear Learner Contingency (SARSA) Human DQN MXDQN Normalized DQN (% Human) Normalized MXDQN (% Human)
Alien 228 939 103 6875 3069 2331.1 43% 32%
Amidar 6 103 184 1676 739.5 829 44% 49%
Assault 222 628 537 1496 3359 2238.8 246% 158%
Asterix 210 987 1332 8503 6012 10433.3 70% 123%
Asteroids 719 907 89 13157 1629 1205.2 7% 4%
Atlantis 12850 62687 853 29028 85641 4119800 450% 25386%
Bank Heist 14 191 67 734 429.7 603.3 58% 82%
Battle Zone 2360 15820 16 37800 26300 33000 68% 86%
Beam Rider 364 929 1743 5775 6846 8957.1 120% 159%
Bowling 23 44 36 155 42.4 57.2 15% 26%
Boxing 0.1 44 10 4 71.8 94.9 1708% 2257%
Breakout 2 5 6 32 401.2 397.9 1327% 1316%
Centipede 2091 8803 4647 11963 8309 4040.2 63% 20%
Chopper Command 811 1582 17 9882 6687 3636.2 65% 31%
Crazy Climber 10781 23411 150 35411 114103 118123.3 420% 436%
Demon Attack 152 521 0 3401 9711 17029.2 294% 519%
Double Dunk -19 -13 -16 -16 -18.1 -10.7 17% 255%
Endure 0 129 159 310 301.8 1421.6 98% 459%
Fishing Derby -92 -90 -85 6 -0.8 25.2 94% 120%
Freeway 0 19 20 30 30.3 30.7 102% 104%
Frostbite 65 217 181 4335 328.3 591.2 6% 12%
Gopher 258 1288 2368 2321 8520 11403.4 400% 540%
Gravitar 173 388 429 2672 306.7 611 5% 18%
H.ER.O. 1027 6459 7295 25763 19950 13783.4 77% 52%
Ice Hockey -11 -10 -3 0.9 -1.6 -3.9 79% 60%
James Bond 29 203 354 407 576.7 536 145% 134%
Kangaroo 52 1622 9 3035 6740 11060 224% 369%
Krull 1598 3372 3341 2395 3805 7904.7 277% 791%
Kung-Fu Master 259 19544 29151 22736 23270 24382.4 102% 107%
Montezuma's Revenge 0 11 259 4367 0 200 0.0% 5%
Ms. Pacman 307 1692 1227 15693 2311 2471.5 13% 14%
Name This Game 2292 2500 2247 4076 7257 10386.5 278% 454%
Pong -21 -19 -17 9 18.9 20 132% 136%
Private Eye 25 684 86 69571 1788 434.8 3% 1%
Dieted 164 614 960 13455 10596 10929.4 79% 81%
River Raid 1339 1904 2650 13513 8316 1451.8 57% 1%
Road Runner 12 68 89 7845 18257 44290.1 233% 565%
Robotank 2 29 12 12 51.6 27.5 509% 261%
Seaquest 68 665 676 20182 5286 28598.9 26% 142%
Space Invaders 148 250 268 1652 1976 1662.1 122% 101%
Star Gunner 664 1070 9 10250 57997 50063.2 598% 515%
Tennis -24 0 0 -9 -2.5 0 143% 160%
Time Pilot 3568 3741 25 5925 5947 6988.9 101% 145%
Tutankham 11 114 98 168 186.7 325 112% 201%
Up and Down 533 3533 2449 9082 8456 16000.3 93% 181%
Venture 0 66 0.6 1188 3830 1151 32% 97%
Video Pinball 16257 16871 19761 17298 42684 211099.5 2539% 18717%
Wizard of Wor 564 1981 37 4757 3393 1343.6 68% 19%
Zazzon 33 3365 21 9173 4977 18261.8 54% 199%

About

MXNet implementation of Deep Q-learning

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 99.1%
  • Other 0.9%