Research Article
An Empirical Investigation of Transfer Effects for Reinforcement Learning
Table 4
Detailed training results of nontransfer and transfer methods to solve sorting 8 numbers for 30 episodes.
| nā=ā8 | | NonTrans_Tr_Steps | Trans_Tr_Steps | Ratio_Tr_Steps | NonTrans_Br_Capacity | Trans_Br_Capacity | Ratio_Br_Capacity |
| 0 | 82101 | 92246 | 0.89 | 0.0624 | 0.0710 | 0.88 | 1 | 77386 | 83553 | 0.93 | 0.0606 | 0.0689 | 0.88 | 2 | 32674 | 17731 | 1.84 | 0.0358 | 0.0452 | 0.79 | 3 | 19818 | 17490 | 1.13 | 0.0340 | 0.0451 | 0.75 | 4 | 24449 | 11835 | 2.07 | 0.0336 | 0.0443 | 0.76 | 5 | 34761 | 27067 | 1.28 | 0.0399 | 0.0487 | 0.82 | 6 | 30299 | 12635 | 2.40 | 0.0348 | 0.0448 | 0.78 | 7 | 26920 | 14774 | 1.82 | 0.0351 | 0.0448 | 0.78 | 8 | 53885 | 44233 | 1.22 | 0.0479 | 0.0546 | 0.88 | 9 | 21150 | 6778 | 3.12 | 0.0293 | 0.0439 | 0.67 | 10 | 56551 | 73505 | 0.77 | 0.0533 | 0.0650 | 0.82 | 11 | 47152 | 43085 | 1.09 | 0.0477 | 0.0544 | 0.88 | 12 | 57590 | 51508 | 1.12 | 0.0505 | 0.0569 | 0.89 | 13 | 21072 | 9521 | 2.21 | 0.0332 | 0.0457 | 0.73 | 14 | 57659 | 36219 | 1.59 | 0.0445 | 0.0500 | 0.89 | 15 | 64347 | 41492 | 1.55 | 0.0523 | 0.0546 | 0.96 | 16 | 31041 | 15146 | 2.05 | 0.0356 | 0.0443 | 0.80 | 17 | 72028 | 72386 | 1.00 | 0.0594 | 0.0678 | 0.88 | 18 | 42305 | 9869 | 4.29 | 0.0393 | 0.0441 | 0.89 | 19 | 29735 | 20833 | 1.43 | 0.0353 | 0.0468 | 0.75 | 20 | 50376 | 46719 | 1.08 | 0.0485 | 0.0559 | 0.87 | 21 | 29481 | 11004 | 2.68 | 0.0452 | 0.0439 | 1.03 | 22 | 37180 | 32229 | 1.15 | 0.0415 | 0.0500 | 0.83 | 23 | 42596 | 25663 | 1.66 | 0.0400 | 0.0473 | 0.85 | 24 | 30466 | 14245 | 2.14 | 0.0337 | 0.0456 | 0.74 | 25 | 62672 | 60769 | 1.03 | 0.0515 | 0.0593 | 0.87 | 26 | 57160 | 55132 | 1.04 | 0.0520 | 0.0538 | 0.97 | 27 | 27488 | 12747 | 2.16 | 0.0466 | 0.0447 | 1.04 | 28 | 28282 | 19243 | 1.47 | 0.0375 | 0.0410 | 0.91 | 29 | 34866 | 26506 | 1.32 | 0.0435 | 0.0477 | 0.91 |
|
|