astier / model-free-episodic-control Goto Github PK

I train your own agent from scratch, but I can't get the same good result as the result in 'agents/Qbert-v0_06-25-06-05-38/results.csv'. Why is that？My result are as follows:

epoch, episodes, frames, reward_sum, reward_avg, reward_max
1,46,101240,31400,683,1050
2,36,99748,34075,947,1125
3,34,101328,35375,1040,1125
4,34,98740,32925,968,1150
5,30,100616,31675,1056,1150
6,31,100416,31025,1001,1150
7,31,100404,31475,1015,1150
8,30,100540,32400,1080,1150
9,29,99980,31750,1095,1150
10,29,97432,31000,1069,1150
11,31,101108,30675,990,1150
12,36,100060,36025,1001,1475
13,32,102100,45200,1412,2000
14,29,96300,45350,1564,2000
15,30,103220,50050,1668,2000
16,28,98972,48825,1744,2025
17,28,99800,48950,1748,2050
18,29,98692,47300,1631,2025
19,28,100180,49900,1782,2050
20,29,100372,48700,1679,2050