pytorchrl.agent.actors.memory_networks package

Submodules

pytorchrl.agent.actors.memory_networks.gru_net module

class pytorchrl.agent.actors.memory_networks.gru_net.GruNet(input_size, output_size=256, num_layers=1, activation=<class 'torch.nn.modules.activation.ReLU'>, dropout=0.0)

Bases: torch.nn.modules.module.Module

Implements a GRU model

forward(inputs, rhs, done)

Forward pass Neural Network

Parameters

• inputs (torch.tensor) – A tensor containing episode observations.

• rhs (torch.tensor) – A tensor representing the recurrent hidden states.

• done (torch.tensor) – A tensor indicating where episodes end.

Returns

• x (torch.tensor) – Output feature map.

• rhs (torch.tensor) – Updated recurrent hidden state.

get_initial_recurrent_state(num_proc)

Returns a tensor of zeros with the expected shape of the model’s rhs.

property num_outputs

Output feature map size (as in np.prod(self.output_shape)).

property recurrent_hidden_state_size

Recurrent hidden state size

training: bool

Module contents

pytorchrl.agent.actors.memory_networks.get_memory_network(name)

Returns model class from name.