05_CartPole-reinforcement-learning_PER_D3QN

SumTree_sample.py

def sample(self, n):
    # Create a minibatch array that will contains the minibatch
    minibatch = []

    b_idx = np.empty((n,), dtype=np.int32)

    # Calculate the priority segment
    # Here, as explained in the paper, we divide the Range[0, ptotal] into n ranges
    priority_segment = self.tree.total_priority / n       # priority segment

    for i in range(n):
        # A value is uniformly sample from each range
        a, b = priority_segment * i, priority_segment * (i + 1)
        value = np.random.uniform(a, b)

        # Experience that correspond to each value is retrieved
        index, priority, data = self.tree.get_leaf(value)

        b_idx[i]= index

        minibatch.append([data[0],data[1],data[2],data[3],data[4]])

    return b_idx, minibatch