policy.py 文件源码

def __init__(self, n_input_channels, n_actions,
                 n_hidden_layers=0, n_hidden_channels=None):
        self.n_input_channels = n_input_channels
        self.n_actions = n_actions
        self.n_hidden_layers = n_hidden_layers
        self.n_hidden_channels = n_hidden_channels

        layers = []
        if n_hidden_layers > 0:
            layers.append(L.Linear(n_input_channels, n_hidden_channels))
            for i in range(n_hidden_layers - 1):
                layers.append(L.Linear(n_hidden_channels, n_hidden_channels))
            layers.append(L.Linear(n_hidden_channels, n_actions))
        else:
            layers.append(L.Linear(n_input_channels, n_actions))

        super(FCSoftmaxPolicy, self).__init__(*layers)