Supporting Tables - Design of Experiments for Reinforcement Learning - page 174

Table C.1 Convergent subregion ranges for the mountain car problem.
Subregion # ʱ mag ʱ ratio ʻ ʳ

10 − 4 ,0 . 010 )

6 [5 . 0 × [1.832, 8.000) [0.000, 0.849) [0.965, 0.993) [0.637, 0.938)

[5 . 0 × 10 − 4 ,0 . 010 )

11 [1.832, 8.000) [0.000, 0.462) [0.993, 1.000) [0.600, 0.657)

10 − 4 ,0 . 010 ) 13 [5 . 0 × [1.832, 2.746) [0.000, 0.849) [0.965, 0.993) [0.938, 0.970)

×

10 − 3 ,0 . 010 ) 25 [9 . 3 [2.746, 3.209) [0.000, 0.679) [0.965, 0.993) [0.948, 0.970)

10 − 3 ,5 . 6 10 − 3 ) 32 [2 . 6 × × [5.069, 7.453) [0.748, 0.952) [0.954, 0.965) [0.826, 0.901)

Table C.2 Kriging model parameters for the mountain car problem.
Trend parameters Covariance parameters

Subregion # Intercept ʱ mag ʱ mag ʻ ʳ ʱ mag ʱ mag ʻ ʳ

− − − −

6 75.523 5.795 0.869 4.829 5.128 31.952 1.196 2.000 2.000 0.574 0.182

11 83.321 − 20.832 2.414 − 4.018 − 8.857 − 9.190 0.153 2.000 2.000 2.000 0.554

13 50.208 − 7.543 1.702 − 9.021 − 3.661 − 0.274 2.000 0.522 0.534 0.147 0.091

25 39.790 − 0.683 1.104 − 3.417 5.044 − 3.102 2.000 2.000 0.494 0.100 2.000

32 48.663 0.890 − 2.288 3.495 − 4.982 − 7.135 2.000 2.000 0.092 2.000 0.058