def test_buffer_sample():
agent_1_buffer = construct_fake_buffer(1)
agent_2_buffer = construct_fake_buffer(2)
update_buffer = AgentBuffer()
agent_1_buffer.resequence_and_append(update_buffer,
batch_size=None,
training_length=2)
agent_2_buffer.resequence_and_append(update_buffer,
batch_size=None,
training_length=2)
# Test non-LSTM
mb = update_buffer.sample_mini_batch(batch_size=4, sequence_length=1)
assert mb.keys() == update_buffer.keys()
assert np.array(mb[BufferKey.CONTINUOUS_ACTION]).shape == (4, 2)
# Test LSTM
# We need to check if we ever get a breaking start - this will maximize the probability
mb = update_buffer.sample_mini_batch(batch_size=20, sequence_length=19)
assert mb.keys() == update_buffer.keys()
# Should only return one sequence
assert np.array(mb[BufferKey.CONTINUOUS_ACTION]).shape == (19, 2)
def test_buffer():
agent_1_buffer = construct_fake_buffer(1)
agent_2_buffer = construct_fake_buffer(2)
agent_3_buffer = construct_fake_buffer(3)
a = agent_1_buffer[ObsUtil.get_name_at(0)].get_batch(batch_size=2,
training_length=1,
sequential=True)
assert_array(np.array(a), np.array([[171, 172, 173], [181, 182, 183]]))
a = agent_2_buffer[ObsUtil.get_name_at(0)].get_batch(batch_size=2,
training_length=3,
sequential=True)
assert_array(
np.array(a),
np.array([
[231, 232, 233],
[241, 242, 243],
[251, 252, 253],
[261, 262, 263],
[271, 272, 273],
[281, 282, 283],
]),
)
a = agent_2_buffer[ObsUtil.get_name_at(0)].get_batch(batch_size=2,
training_length=3,
sequential=False)
assert_array(
np.array(a),
np.array([
[251, 252, 253],
[261, 262, 263],
[271, 272, 273],
[261, 262, 263],
[271, 272, 273],
[281, 282, 283],
]),
)
agent_1_buffer.reset_agent()
assert agent_1_buffer.num_experiences == 0
update_buffer = AgentBuffer()
agent_2_buffer.resequence_and_append(update_buffer,
batch_size=None,
training_length=2)
agent_3_buffer.resequence_and_append(update_buffer,
batch_size=None,
training_length=2)
assert len(update_buffer[BufferKey.CONTINUOUS_ACTION]) == 20
assert np.array(update_buffer[BufferKey.CONTINUOUS_ACTION]).shape == (20,
2)
c = update_buffer.make_mini_batch(start=0, end=1)
assert c.keys() == update_buffer.keys()
assert np.array(c[BufferKey.CONTINUOUS_ACTION]).shape == (1, 2)
def test_buffer():
b = construct_fake_processing_buffer()
a = b[1]["vector_observation"].get_batch(batch_size=2,
training_length=1,
sequential=True)
assert_array(np.array(a), np.array([[171, 172, 173], [181, 182, 183]]))
a = b[2]["vector_observation"].get_batch(batch_size=2,
training_length=3,
sequential=True)
assert_array(
np.array(a),
np.array([
[231, 232, 233],
[241, 242, 243],
[251, 252, 253],
[261, 262, 263],
[271, 272, 273],
[281, 282, 283],
]),
)
a = b[2]["vector_observation"].get_batch(batch_size=2,
training_length=3,
sequential=False)
assert_array(
np.array(a),
np.array([
[251, 252, 253],
[261, 262, 263],
[271, 272, 273],
[261, 262, 263],
[271, 272, 273],
[281, 282, 283],
]),
)
b[4].reset_agent()
assert len(b[4]) == 0
update_buffer = AgentBuffer()
b.append_to_update_buffer(update_buffer,
3,
batch_size=None,
training_length=2)
b.append_to_update_buffer(update_buffer,
2,
batch_size=None,
training_length=2)
assert len(update_buffer["action"]) == 20
assert np.array(update_buffer["action"]).shape == (20, 2)
c = update_buffer.make_mini_batch(start=0, end=1)
assert c.keys() == update_buffer.keys()
assert np.array(c["action"]).shape == (1, 2)
def test_buffer_sample():
b = construct_fake_processing_buffer()
update_buffer = AgentBuffer()
b.append_to_update_buffer(update_buffer,
3,
batch_size=None,
training_length=2)
b.append_to_update_buffer(update_buffer,
2,
batch_size=None,
training_length=2)
# Test non-LSTM
mb = update_buffer.sample_mini_batch(batch_size=4, sequence_length=1)
assert mb.keys() == update_buffer.keys()
assert np.array(mb["action"]).shape == (4, 2)
# Test LSTM
# We need to check if we ever get a breaking start - this will maximize the probability
mb = update_buffer.sample_mini_batch(batch_size=20, sequence_length=19)
assert mb.keys() == update_buffer.keys()
# Should only return one sequence
assert np.array(mb["action"]).shape == (19, 2)
def test_buffer():
agent_1_buffer = construct_fake_buffer(1)
agent_2_buffer = construct_fake_buffer(2)
agent_3_buffer = construct_fake_buffer(3)
# Test get_batch
a = agent_1_buffer[ObsUtil.get_name_at(0)].get_batch(batch_size=2,
training_length=1,
sequential=True)
assert_array(
np.array(a),
np.array([[171, 172, 173], [181, 182, 183]], dtype=np.float32))
# Test get_batch
a = agent_2_buffer[ObsUtil.get_name_at(0)].get_batch(batch_size=2,
training_length=3,
sequential=True)
assert_array(
np.array(a),
np.array(
[
[231, 232, 233],
[241, 242, 243],
[251, 252, 253],
[261, 262, 263],
[271, 272, 273],
[281, 282, 283],
],
dtype=np.float32,
),
)
a = agent_2_buffer[ObsUtil.get_name_at(0)].get_batch(batch_size=2,
training_length=3,
sequential=False)
assert_array(
np.array(a),
np.array([
[251, 252, 253],
[261, 262, 263],
[271, 272, 273],
[261, 262, 263],
[271, 272, 273],
[281, 282, 283],
]),
)
# Test padding
a = agent_2_buffer[ObsUtil.get_name_at(0)].get_batch(batch_size=None,
training_length=4,
sequential=True)
assert_array(
np.array(a),
np.array([
[201, 202, 203],
[211, 212, 213],
[221, 222, 223],
[231, 232, 233],
[241, 242, 243],
[251, 252, 253],
[261, 262, 263],
[271, 272, 273],
[281, 282, 283],
[0, 0, 0],
[0, 0, 0],
[0, 0, 0],
]),
)
# Test group entries return Lists of Lists. Make sure to pad properly!
a = agent_2_buffer[BufferKey.GROUP_CONTINUOUS_ACTION].get_batch(
batch_size=None, training_length=4, sequential=True)
for _group_entry in a[:-3]:
assert len(_group_entry) == 3
for _group_entry in a[-3:]:
assert len(_group_entry) == 0
agent_1_buffer.reset_agent()
assert agent_1_buffer.num_experiences == 0
update_buffer = AgentBuffer()
agent_2_buffer.resequence_and_append(update_buffer,
batch_size=None,
training_length=2)
agent_3_buffer.resequence_and_append(update_buffer,
batch_size=None,
training_length=2)
assert len(update_buffer[BufferKey.CONTINUOUS_ACTION]) == 20
assert np.array(update_buffer[BufferKey.CONTINUOUS_ACTION]).shape == (20,
2)
c = update_buffer.make_mini_batch(start=0, end=1)
assert c.keys() == update_buffer.keys()
# Make sure the values of c are AgentBufferField
for val in c.values():
assert isinstance(val, AgentBufferField)
assert np.array(c[BufferKey.CONTINUOUS_ACTION]).shape == (1, 2)
