Sampling method in D4PG in alignment to required data type.

sampling_method_in_d4pg.py

# Excerpt of replay memory object.
class ReplayBuffer:
    """Fixed-size buffer to store experience tuples."""
    
    def sample2(self, device=device):
    """Randomly sample a batch of experiences from memory."""
    experiences = random.sample(self.memory, k=self.batch_size)

    states, actions, rewards, next_states, dones = [], [], [], [], []

    for exp in experiences:
        states.append(exp.state.squeeze(0))
        actions.append(exp.action.squeeze(0))
        rewards.append(exp.reward)
        dones.append(exp.done)
        next_states.append(exp.next_state.squeeze(0))

    states_v = torch.Tensor(np.array(states, dtype=np.float32)).to(device)
    actions_v = torch.Tensor(np.array(actions, dtype=np.float32)).to(device)
    rewards_v = torch.Tensor(np.array(rewards, dtype=np.float32)).to(device)
    next_states_v = torch.Tensor(np.array(next_states, dtype=np.float32)).to(device)
    dones_v = torch.ByteTensor(dones).to(device)

    return states_v, actions_v, rewards_v, next_states_v, dones_v