Where is numpy.object_ from?

[fulacse]

from marllib import marl

env = marl.make_env(environment_name="sisl", map_name="pursuit")

mappo = marl.algos.mappo(hyperparam_source="common")

model = marl.build_model(env, mappo, {
"conv_layer": 3,
"out_channel_layer_0": 64,
"kernel_size_layer_0": 3,
"stride_layer_0": 1,
"padding_layer_0": 0,
"pool_size_layer_0": 1,
"out_channel_layer_1": 128,
"kernel_size_layer_1": 3,
"stride_layer_1": 1,
"padding_layer_1": 0,
"pool_size_layer_1": 1,
"out_channel_layer_2": 256,
"kernel_size_layer_2": 3,
"stride_layer_2": 1,
"padding_layer_2": 0,
"pool_size_layer_2": 1,
"hidden_state_size": 512,
"core_arch": "mlp",
})

mappo.fit(env, model, stop={'timesteps_total': 1000000}, share_policy='all', checkpoint_freq=10, checkpoint_end=True,
local_mode=False, num_gpus=1, num_workers=5)
error.txt

[fulacse]

and I don't want pool layer, how do I remove it?

[Aequatio-Space]

For the "numpy.object_" problem, try checking ray-project/ray#26863, you may find the solution in commits.
Apply a patch to the ray library will solve the problem.

Pooling layer is hardcoded in here, if you want to get rid out of it, some modification of the library may needed.

MARLlib/marllib/marl/models/zoo/encoder/base_encoder.py

Lines 79 to 80 in 368c617

	pool_f = nn.MaxPool2d(kernel_size=self.custom_config["model_arch_args"]["pool_size_layer_{}".format(i)])
	layers.append(pool_f)

[fulacse]

For the "numpy.object_" problem, try checking ray-project/ray#26863, you may find the solution in commits. Apply a patch to the ray library will solve the problem.

Pooling layer is hardcoded in here, if you want to get rid out of it, some modification of the library may needed.

MARLlib/marllib/marl/models/zoo/encoder/base_encoder.py

Lines 79 to 80 in 368c617

	pool_f = nn.MaxPool2d(kernel_size=self.custom_config["model_arch_args"]["pool_size_layer_{}".format(i)])
	layers.append(pool_f)

Thanks, is probleme of pythorch version. And I think it don't change anything if pool_size = 1