minihack/agent/polybeast/models/base.py [31:54]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
class NetHackNet(nn.Module):
    AgentOutput = collections.namedtuple(
        "AgentOutput", "action policy_logits baseline"
    )

    def __init__(self):
        super(NetHackNet, self).__init__()

        self.register_buffer("reward_sum", torch.zeros(()))
        self.register_buffer("reward_m2", torch.zeros(()))
        self.register_buffer("reward_count", torch.zeros(()).fill_(1e-8))

    def forward(self, inputs, core_state):
        raise NotImplementedError

    def initial_state(self, batch_size=1):
        return ()

    def prepare_input(self, inputs):
        # -- [T x B x H x W]
        glyphs = inputs["glyphs"]

        # -- [T x B x F]
        features = inputs["blstats"]
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



minihack/agent/rllib/models.py [67:90]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
class NetHackNet(nn.Module):
    AgentOutput = collections.namedtuple(
        "AgentOutput", "action policy_logits baseline"
    )

    def __init__(self):
        super(NetHackNet, self).__init__()

        self.register_buffer("reward_sum", torch.zeros(()))
        self.register_buffer("reward_m2", torch.zeros(()))
        self.register_buffer("reward_count", torch.zeros(()).fill_(1e-8))

    def forward(self, inputs, core_state):
        raise NotImplementedError

    def initial_state(self, batch_size=1):
        return ()

    def prepare_input(self, inputs):
        # -- [B x H x W]
        glyphs = inputs["glyphs"]

        # -- [B x F]
        features = inputs["blstats"]
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -