def __init__(self, size, alpha):
"""Create Prioritized Replay buffer.
Parameters
----------
size: int
Max number of transitions to store in the buffer. When the buffer
overflows the old memories are dropped.
alpha: float
how much prioritization is used
(0 - no prioritization, 1 - full prioritization)
See Also
--------
ReplayBuffer.__init__
"""
super(PrioritizedReplayBuffer, self).__init__(size)
assert alpha > 0
self._alpha = alpha
it_capacity = 1
while it_capacity < size:
it_capacity *= 2
self._it_sum = SumSegmentTree(it_capacity)
self._it_min = MinSegmentTree(it_capacity)
self._max_priority = 1.0
self._prio_change_stats = WindowStat("reprio", 1000)
def __init__(
self,
size,
alpha,
# added for dynamic experience replay
permanent_data_length=0,
human_data=None,
dynamic_experience_replay=False,
demonstration_zone_percentage=0):
"""Create Prioritized Replay buffer.
Parameters
----------
size: int
Max number of transitions to store in the buffer. When the buffer
overflows the old memories are dropped.
alpha: float
how much prioritization is used
(0 - no prioritization, 1 - full prioritization)
See Also
--------
ReplayBuffer.__init__
"""
# added for dynamic experience replay
super(PrioritizedReplayBuffer, self).__init__(size,
permanent_data_length)
assert alpha > 0
self._alpha = alpha
it_capacity = 1
while it_capacity < size:
it_capacity *= 2
self._it_sum = SumSegmentTree(it_capacity)
self._it_min = MinSegmentTree(it_capacity)
self._max_priority = 1.0
self._prio_change_stats = WindowStat("reprio", 1000)
# added for dynamic experience replay
self._permanent_data_length = permanent_data_length
self._demo_index = permanent_data_length
if dynamic_experience_replay and demonstration_zone_percentage > 0:
self._max_demo_size = math.ceil(
size * demonstration_zone_percentage
) # assign the size for demonstration zone
assert self._max_demo_size > 10000, "demonstration zone is too small, please increase buffer size"
prGreen("the maximum demonstration zone size: {}".format(
self._max_demo_size))
if self._permanent_data_length > 0:
# insert human demos to the buffer with full prioritization
for transition in human_data:
self.add(transition[0], transition[1], transition[2],
transition[3], transition[4], 1)
def test_tree_set_overlap(self):
tree = SumSegmentTree(4)
tree[2] = 1.0
tree[2] = 3.0
assert np.isclose(tree.sum(), 3.0)
assert np.isclose(tree.sum(2, 3), 3.0)
assert np.isclose(tree.sum(2, -1), 3.0)
assert np.isclose(tree.sum(2, 4), 3.0)
assert np.isclose(tree.sum(1, 2), 0.0)
def test_tree_set(self):
tree = SumSegmentTree(4)
tree[2] = 1.0
tree[3] = 3.0
assert np.isclose(tree.sum(), 4.0)
assert np.isclose(tree.sum(0, 2), 0.0)
assert np.isclose(tree.sum(0, 3), 1.0)
assert np.isclose(tree.sum(2, 3), 1.0)
assert np.isclose(tree.sum(2, -1), 1.0)
assert np.isclose(tree.sum(2, 4), 4.0)
def test_prefixsum_idx(self):
tree = SumSegmentTree(4)
tree[2] = 1.0
tree[3] = 3.0
assert tree.find_prefixsum_idx(0.0) == 2
assert tree.find_prefixsum_idx(0.5) == 2
assert tree.find_prefixsum_idx(0.99) == 2
assert tree.find_prefixsum_idx(1.01) == 3
assert tree.find_prefixsum_idx(3.00) == 3
assert tree.find_prefixsum_idx(4.00) == 3
def test_prefixsum_idx2(self):
tree = SumSegmentTree(4)
tree[0] = 0.5
tree[1] = 1.0
tree[2] = 1.0
tree[3] = 3.0
assert tree.find_prefixsum_idx(0.00) == 0
assert tree.find_prefixsum_idx(0.55) == 1
assert tree.find_prefixsum_idx(0.99) == 1
assert tree.find_prefixsum_idx(1.51) == 2
assert tree.find_prefixsum_idx(3.00) == 3
assert tree.find_prefixsum_idx(5.50) == 3