Inputs

[Yiquan-lol]

hello
Is the input size [168,168,3] or [84,84,3] in https://github.com/praveen-palanisamy/macad-agents/blob/master/src/macad_agents/rllib/impala_multiagent_shared_weights.py？
In addition, conv_filter is [[32, [8], 4], [64, [4], 2], [64, [3, 3], 1)]?

[praveen-palanisamy]

The observation for the HomoNcomIndePOIntrxMASS3CTWN3 MACAD-Gym environment as per the environment config is [168, 168, 3]. This is the shape of the observation sent to the Impala_multiagent_shared_weights.py agent trainer script. This script resizes the observations to [84, 84, 3] before feeding the agent.
The Impala agent's CNN policy IIRC, uses the following conv filter configuration:

[[16, [8], 4],
[32, [4], 2],
[256, [11], 1]]

You can customize the shape of the Agent's input by changing these lines:

macad-agents/src/macad_agents/rllib/impala_multiagent_shared_weights.py

Lines 71 to 82 in b2726b3

	class ImagePreproc(Preprocessor):
	def _init_shape(self, obs_space, options):
	self.shape = (84, 84, 3) # Adjust third dim if stacking frames
	return self.shape
	def transform(self, observation):
	observation = cv2.resize(observation, (self.shape[0], self.shape[1]))
	return observation
	ModelCatalog.register_custom_preprocessor("sq_im_84", ImagePreproc)

You can also customize the filters used by un-commenting and modifying these lines:

macad-agents/src/macad_agents/rllib/impala_multiagent_shared_weights.py

Lines 107 to 109 in b2726b3

	# conv_filters to be used with the custom CNN model.
	# "conv_filters": [[16, [4, 4], 2], [32, [3, 3], 2],
	# [16, [3, 3], 2]]