Research Article
An Empirical Investigation of Transfer Effects for Reinforcement Learning
Table 1
Detailed training results of nontransfer and transfer methods to solve sorting 5 numbers for 30 episodes.
| nā=ā5 | | NonTrans_Tr_Steps | Trans_Tr_Steps | Ratio_Tr_Steps | NonTrans_Br_Capacity | Trans_Br_Capacity | Ratio_Br_Capacity |
| 0 | 167 | 20 | 8.35 | 0.1983 | 0.1500 | 1.32 | 1 | 215 | 94 | 2.29 | 0.1808 | 0.2342 | 0.77 | 2 | 964 | 365 | 2.64 | 0.2808 | 0.2142 | 1.31 | 3 | 207 | 42 | 4.93 | 0.2025 | 0.1533 | 1.32 | 4 | 94 | 120 | 0.78 | 0.1817 | 0.1717 | 1.06 | 5 | 189 | 110 | 1.72 | 0.1633 | 0.1825 | 0.89 | 6 | 361 | 22 | 16.41 | 0.2092 | 0.1783 | 1.17 | 7 | 146 | 22 | 6.64 | 0.1675 | 0.1642 | 1.02 | 8 | 94 | 335 | 0.28 | 0.1892 | 0.2083 | 0.91 | 9 | 382 | 118 | 3.24 | 0.2208 | 0.1742 | 1.27 | 10 | 230 | 118 | 1.95 | 0.1817 | 0.1850 | 0.98 | 11 | 78 | 32 | 2.44 | 0.1225 | 0.1775 | 0.69 | 12 | 276 | 48 | 5.75 | 0.1825 | 0.1517 | 1.20 | 13 | 130 | 60 | 2.17 | 0.2067 | 0.1525 | 1.36 | 14 | 320 | 64 | 5.00 | 0.2242 | 0.1658 | 1.35 | 15 | 241 | 96 | 2.51 | 0.1875 | 0.1883 | 1.00 | 16 | 286 | 38 | 7.53 | 0.2075 | 0.1633 | 1.27 | 17 | 246 | 128 | 1.92 | 0.2058 | 0.2042 | 1.01 | 18 | 140 | 114 | 1.23 | 0.1867 | 0.1567 | 1.19 | 19 | 249 | 46 | 5.41 | 0.3217 | 0.1575 | 2.04 | 20 | 130 | 532 | 0.24 | 0.1775 | 0.2183 | 0.81 | 21 | 154 | 10 | 15.40 | 0.0983 | 0.1000 | 0.98 | 22 | 10 | 36 | 0.28 | 0.0558 | 0.1042 | 0.54 | 23 | 456 | 175 | 2.61 | 0.2242 | 0.1842 | 1.22 | 24 | 101 | 12 | 8.42 | 0.0717 | 0.1175 | 0.61 | 25 | 400 | 84 | 4.76 | 0.1775 | 0.1183 | 1.50 | 26 | 117 | 109 | 1.07 | 0.0983 | 0.1283 | 0.77 | 27 | 241 | 113 | 2.13 | 0.1433 | 0.1633 | 0.88 | 28 | 184 | 50 | 3.68 | 0.2008 | 0.1858 | 1.08 | 29 | 102 | 56 | 1.82 | 0.1583 | 0.1617 | 0.98 |
|
|