Question: How to create a model which takes two inputs?

[HadiSDev]

Hi, I have an environment which can return an image x and a feature vector f

How can I construct my CNN/FN such that my model can take them both: Model(x, f)

My current Q-CNN is just with the image:

class QFunction(chainer.Chain):
    def __init__(self, n_history=1, n_action=3):
        super().__init__(
            l1=L.Convolution2D(n_history, 32, ksize=3, stride=2, nobias=False),
            l2=L.Convolution2D(32, 64, ksize=3, stride=2, nobias=False),
            l3=L.Convolution2D(64, 64, ksize=3, stride=2, nobias=False),
            l4=L.Linear(256, 128),
            out=L.Linear(n_action)
        )

    def __call__(self, x, test=False):
        h1 = F.relu(self.l1(x))
        h2 = F.relu(self.l2(h1))
        h3 = F.relu(self.l3(h2))
        h4 = F.relu(self.l4(h3))
        h5 = self.out(h4)
        return chainerrl.action_value.DiscreteActionValue(h5)

[muupan]

You can define your Q-function with __call__ taking a tuple (x, f) as input.

See the grasping example: https://github.com/chainer/chainerrl/blob/master/examples/grasping/train_dqn_batch_grasping.py#L94