RuntimeError: Could not infer dtype of numpy.int64

RuntimeError: Could not infer dtype of numpy.int64

问题：再写强化学习代码时，遇到该错误，代码如下，对于传的参数max_action=2.0，使用
self.action_scale = torch.tensor(max_action)会报如上错误，主要是max_action变为了numpy.int64，所以不能用torch.tensor()

class Actor(nn.Module):
    # def __init__(self, action_dim, state_dim, log_std_min=-20, log_std_max=2, max_action=None):  # 这样会把min, max变为numpy.float
    def __init__(self, action_dim, state_dim, env, max_action=None):
        super(Actor, self).__init__()
        # self.min_log_std = log_std_min
        # self.max_log_std = log_std_max
        self.f1 = nn.Linear(state_dim, 200)
        self.f2 = nn.Linear(200, 200)
        self.mean = nn.Linear(200, action_dim)
        self.log_std = nn.Linear(200, action_dim)

        self.mean.weight.data.uniform_(-INIT_W, INIT_W)
        self.mean.bias.data.uniform_(-INIT_W, INIT_W)

        self.log_std.weight.data.uniform_(-INIT_W, INIT_W)
        self.log_std.bias.data.uniform_(-INIT_W, INIT_W)

        # action scaling
        if max_action == None:
            self.action_scale = torch.tensor(1.)
            self.action_bias = torch.tensor(0.)
        else:
            self.action_scale = torch.tensor(max_action)
            self.action_bias = torch.tensor(0.)

解决：使用torch.tensor()的时候加参数dtype=torch.float就可以。

self.action_scale = torch.tensor(float(max_action), dtype=torch.float)
self.action_bias = torch.tensor(0., dtype=torch.float)