Puzzle with this line

[sl950313]

pensieve/multi_video_sim/multi_agent.py

Line 284 in 1120bb1

state = [np.zeros((S_INFO, S_LEN))]

state is array with shape (S_INFO, S_LEN), s_batch is container of previous states with shape (1, 6, 8). why the state is created with shape=(1, 6, 8) ?
Please help, thanks!

[hongzimao]

I think this is just to make the initial state compatible with other states for the np.roll step in

pensieve/multi_video_sim/multi_agent.py

Line 289 in 1120bb1

state = np.roll(state, -1, axis=1)

. You can pdb the code and examine the shapes of those variables. Hope this helps.