Griddly rollout example about griddly HOT 3 CLOSED

I dont have any specific examples for this.

From the size of your input vector and the trained weights, it looks like you might be training using VECTOR observers and then rolling out in the model with either BLOCK_2D or SPRITE_2D observers.

make sure you set player_observer_type=gd.ObserverType.VECTOR in the environment where you are rolling out.

from griddly.

Thank you ! I made it working now. As you have mentioned in your document "By default the agent sees a VECTOR view of the environment. This view is passed to a Global Average Pooling Agent to produce the policy."
In fact, there still 2 points. Obs is of type unit8 so I changed it to obs = obs.astype('float64')
In addition, in agent code you are using: permute(0, 3, 1, 2), so I changed my input to obs = np.moveaxis(obs, 0, 2).
Is this make sense to you ?
Best

from griddly.

Yeah this makes sense, glad you got it working!

from griddly.