DDPG init

[PkuDavidGuan]

I think the actor network and the actor target network should have the same parameters before the training, so I think the function initialize should be called in DDPG.init. https://github.com/megvii-rl/pytorch-gym/blob/6c9223e6e3fc5629580eec5c3f0a262a17caa5f3/DDPG_from_baselines/pytorch_base/ddpg.py#L19