def test_sum_priority(self):
sumTree = SumTree(10, [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8),
Item("A", 5), Item("A", 8), Item("A", 7), Item("A", -1)])
p = 0
for leaf in sumTree.get_leaves():
p += leaf.get_priority()
self.assertAlmostEqual(p, sumTree.get_sum_priority(), places=8)
def test_TD_order(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5), Item("A", 8), Item("A", 7), Item("A", -1)]
sumTree = SumTree(10, items)
curr = None
for leaf in sumTree.get_leaves():
if curr is None:
curr = abs(leaf.TD_error)
continue
self.assertGreaterEqual(curr, abs(leaf.TD_error))
curr = abs(leaf.TD_error)
def test_remove_add_items(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5)]
sumTree = SumTree(7, items)
total = sumTree.get_sum_priority()
x,y,z = Item("X", 99), Item("Y", 98), Item("Z", 97)
sumTree.add_item(x)
sumTree.add_item(y)
sumTree.add_item(z)
self.assertListEqual(list(sumTree.get_leaves())[:3], [z, y, x])
def test_rank_order(self):
sumTree = SumTree(10, [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8),
Item("A", 5), Item("A", 8), Item("A", 7), Item("A", -1)])
curr = None
for leaf in sumTree.get_leaves():
if curr is None:
curr = leaf.get_rank()
continue
self.assertLess(curr, leaf.get_rank())
curr = leaf.get_rank()
def test_priority_order(self):
sumTree = SumTree(10, [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8),
Item("A", 5), Item("A", 8), Item("A", 7), Item("A", -1)])
curr = None
for leaf in sumTree.get_leaves():
if curr is None:
curr = leaf.get_priority()
continue
self.assertGreaterEqual(curr, leaf.get_priority())
curr = leaf.get_priority()
def test_adding_speed(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5)]
sumTree = SumTree(1000000, items)
print("Adding Speed")
for x in items:
start = time.perf_counter()
sumTree.add_item(x)
end = time.perf_counter()
elapsed_time = end - start
print("Time:", elapsed_time)
self.assertLessEqual(elapsed_time, 0.3)
print("\n")
def test_adding_priority(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5)]
sumTree = SumTree(7, [])
for x in items:
sumTree.add_item(x)
curr = None
for leaf in sumTree.get_leaves():
if curr is None:
curr = leaf.get_priority()
continue
self.assertGreaterEqual(curr, leaf.get_priority())
curr = leaf.get_priority()
def test_minibatch_speed(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5)]
sumTree = SumTree(1000000, items)
# start = time.perf_counter()
# sumTree.get_minibatch(32)
# end = time.perf_counter()
start2 = time.perf_counter()
sumTree.get_minibatch2(32)
end2 = time.perf_counter()
## print("MinB ", end-start)
print("Minibatch Speed")
print("Time: ", end2-start2)
print("\n")
self.assertLessEqual(end2-start2, 0.003)
def test_changing_priority(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5)]
sumTree = SumTree(7, items)
total = sumTree.get_sum_priority()
x,y,z = Item("X", 99), Item("Y", 98), Item("Z", 97)
sumTree.add_item(x)
sumTree.add_item(y)
sumTree.add_item(z)
curr = None
for leaf in sumTree.get_leaves():
if curr is None:
curr = leaf.get_priority()
continue
self.assertGreaterEqual(curr, leaf.get_priority())
curr = leaf.get_priority()
def test_minibatch3(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5)]
sumTree = SumTree(1000000, items)
sumTree.get_minibatch3(32)
def test_save_load(self):
items = [Item("A", 6), Item("A", 3), Item("A", 999), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", None), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 68), Item("A", 5),
Item("A", 16), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 89), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 33), Item("A", 1), Item("A", 70), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", -89), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 68), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 15),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 15)]
def save_data(frame_number, my_replay_memory, log_list, sess, test=False):
if test:
with open('memory.pkl', 'wb') as output:
pickle.dump(my_replay_memory, output, pickle.HIGHEST_PROTOCOL)
return
def load_data(sess, test=False):
if test:
with open('memory.pkl', 'rb') as input:
my_replay_memory = pickle.load(input)
return my_replay_memory
sumTree = SumTree(1000000, items)
MEM = PEReplayMemory()
MEM.tree = sumTree
save_data(0, MEM, [], None, test=True)
my_replay_memory = load_data(None, test=True)
self.assertEqual(my_replay_memory.tree.tree, MEM.tree.tree)
def test_tree_sort(self):
items = [Item("A", 6), Item("A", 3), Item("A", 999), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", None), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 68), Item("A", 5),
Item("A", 16), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 89), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 33), Item("A", 1), Item("A", 70), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", -89), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 68), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 15),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 5),
Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8), Item("A", 15)]
sumTree = SumTree(1000000, items)
PEReplayMemory.tree = sumTree
PER = PEReplayMemory()
x = np.zeros((84, 84, 4))
PER.add_experience(0, x, 0, 0, False)
PER.tree.add_item(Item("None", None))
PER.tree.add_item(Item("Skipped"))
PER.sort_tree()
curr = None
for leaf in sumTree.get_leaves():
if curr is None:
curr = abs(leaf.TD_error)
continue
self.assertGreaterEqual(curr, abs(leaf.TD_error))
curr = abs(leaf.TD_error)
def test_leaf_internal_idx(self):
sumTree = SumTree(10, [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9), Item("A", 8),
Item("A", 5), Item("A", 8), Item("A", 7), Item("A", -1)])
for idx in range(sumTree.get_num_leaves()):
self.assertEqual(sumTree.tree[idx].idx, idx - sumTree.idx_shift)
def test_get_leaves(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9),
Item("A", 8), Item("A", 5), Item("A", 8), Item("A", 7), Item("A", -1)]
sumTree = SumTree(10, items)
self.assertEqual(sumTree.get_leaves(), deque(items, maxlen=10))
def test_num_leaves(self):
items = [Item("A", 6), Item("A", 3), Item("A", 1), Item("A", 0), Item("A", 9),
Item("A", 8), Item("A", 5), Item("A", 8), Item("A", 7), Item("A", -1)]
sumTree = SumTree(10, items)
self.assertEqual(sumTree.get_num_leaves(), len(items))