def test_per_buffer_reset_alpha():
# Assign
per_buffer = PERBuffer(10, 10, alpha=0.1)
for _ in range(30):
per_buffer.add(reward=np.random.randint(0, 1e5),
priority=np.random.random())
# Act
old_experiences = per_buffer.sample()
per_buffer.reset_alpha(0.5)
new_experiences = per_buffer.sample()
# Assert
assert old_experiences is not None and new_experiences is not None
old_index, new_index = np.array(old_experiences['index']), np.array(
new_experiences['index'])
old_weight, new_weight = np.array(old_experiences['weight']), np.array(
new_experiences['weight'])
old_reward, new_reward = np.array(old_experiences['reward']), np.array(
new_experiences['reward'])
old_sort, new_sort = np.argsort(old_index), np.argsort(new_index)
assert all([
i1 == i2 for (i1, i2) in zip(old_index[old_sort], new_index[new_sort])
])
assert all([
w1 != w2
for (w1, w2) in zip(old_weight[old_sort], new_weight[new_sort])
])
assert all([
r1 == r2
for (r1, r2) in zip(old_reward[old_sort], new_reward[new_sort])
])
def test_per_buffer_reset_alpha():
# Assign
per_buffer = PERBuffer(10, 10, alpha=0.1)
for _ in range(30):
per_buffer.add(reward=np.random.randint(0, 1e5),
priority=np.random.random())
# Act
old_experiences = per_buffer.sample_list()
per_buffer.reset_alpha(0.5)
new_experiences = per_buffer.sample_list()
# Assert
assert old_experiences is not None and new_experiences is not None
sorted_new_experiences = sorted(new_experiences, key=lambda k: k.index)
sorted_old_experiences = sorted(old_experiences, key=lambda k: k.index)
for (new_sample, old_sample) in zip(sorted_new_experiences,
sorted_old_experiences):
assert new_sample.index == old_sample.index
assert new_sample.weight != old_sample.weight
assert new_sample.reward == old_sample.reward