def test_n_step_td_step_adjust(self, context: Context) -> None:
"""Test adjusting steps taken on reward signal propagation."""
predictor = TemporalDifference(step=1)
predictor._temperature = 1.0
predictor._steps_taken = 1
package_tuple = ("tensorflow", "2.3.1", "https://pypi.org/simple")
state = State()
state.add_resolved_dependency(package_tuple)
with predictor.assigned_context(context):
predictor.set_reward_signal(state, package_tuple, 0.33)
assert predictor._policy.get(package_tuple) == [0.33, 1]
assert predictor._steps_taken == 0
def test_set_reward_signal_nan_inf(self, float_case: float) -> None:
"""Test (not) keeping the reward signal for nan/inf."""
predictor = TemporalDifference()
state = State()
state.add_resolved_dependency(
("tensorflow", "2.3.0", "https://pypi.org/simple"))
state.add_resolved_dependency(
("flask", "0.12", "https://pypi.org/simple"))
state.add_unresolved_dependency(
("termial-random", "0.0.2", "https://pypi.org/simple"))
predictor._policy = {
("flask", "0.12", "https://pypi.org/simple"): [0.2, 1],
}
predictor._steps_taken = 2
predictor._steps_reward = 1.2
predictor._next_state = state
assert (predictor.set_reward_signal(
state, ("tensorflow", "2.0.0", "https://pypi.org/simple"),
float_case) is None)
assert predictor._policy == {
("flask", "0.12", "https://pypi.org/simple"): [1.4, 2],
("tensorflow", "2.3.0", "https://pypi.org/simple"): [1.2, 1],
}
assert predictor._steps_taken == 0
assert predictor._steps_reward == 0.0
assert predictor._next_state is None
def test_set_reward_signal_unseen(self) -> None:
"""Test keeping the reward signal for an unseen step."""
reward = 42.24
package_tuple = ("tensorflow", "2.0.0", "https://thoth-station.ninja")
state = flexmock()
state.should_receive("iter_resolved_dependencies").and_return([package_tuple]).once()
predictor = TemporalDifference()
predictor._policy = {
("numpy", "1.0.0", "https://pypi.org/simple"): [30.30, 92],
}
predictor._steps_taken = 1
predictor.set_reward_signal(state, None, reward)
assert predictor._policy == {
package_tuple: [42.24, 1],
("numpy", "1.0.0", "https://pypi.org/simple"): [30.30, 92],
}