DQN

[stegben]

Deep Q learning

Followed by the pseudo-code of algorithm 1 in this paper, using decay-epsilon greedy to select actions.

[stegben]

Uh.. plz don't merge right now, let me try to get a better default parameters.

[stegben]

Done.

[2017-06-18 14:15:18,993] Making new env: CartPole-v0
Episode 0	Last length:    23	Average length: 2.30
Episode 10	Last length:    13	Average length: 31.30
Episode 20	Last length:    31	Average length: 17.40
Episode 30	Last length:    28	Average length: 18.70
Episode 40	Last length:    13	Average length: 17.90
Episode 50	Last length:    10	Average length: 13.70
Episode 60	Last length:    13	Average length: 16.00
Episode 70	Last length:    15	Average length: 19.70
Episode 80	Last length:    11	Average length: 15.00
Episode 90	Last length:    10	Average length: 14.30
Episode 100	Last length:    13	Average length: 12.90
Episode 110	Last length:    11	Average length: 13.60
Episode 120	Last length:    27	Average length: 14.70
Episode 130	Last length:    17	Average length: 29.50
Episode 140	Last length:    21	Average length: 31.10
Episode 150	Last length:    64	Average length: 45.80
Episode 160	Last length:   142	Average length: 97.60
Episode 170	Last length:   196	Average length: 163.60
Episode 180	Last length:   144	Average length: 184.80
Episode 190	Last length:   199	Average length: 196.70
Episode 200	Last length:   199	Average length: 194.70
Episode 210	Last length:   171	Average length: 193.30
Episode 220	Last length:   199	Average length: 198.10

[stegben]

Oh, just found that there're tutorial in the official page