def test_interpretation_plot_q_dqn_returns():
data = MDPDataBunch.from_env('maze-random-5x5-v0', render='human')
model = DQN(data)
learn = AgentLearner(data, model)
learn.fit(5)
interp = AgentInterpretationAlpha(learn)
interp.plot_heatmapped_episode(2)
def test_fit_function_dqn():
data = MDPDataBunch.from_env('maze-random-5x5-v0',
render='human',
max_steps=1000)
model = DQN(data, memory=PriorityExperienceReplay(1000))
learn = AgentLearner(data, model)
learn.fit(5)
def test_priority_experience_replay():
data = MDPDataBunch.from_env('maze-random-5x5-v0',
render='human',
max_steps=1000)
model = FixedTargetDQN(data, memory=PriorityExperienceReplay(1000))
learn = AgentLearner(data, model)
learn.fit(5)
def test_double_dueling_dqn_model_maze():
data = MDPDataBunch.from_env('maze-random-5x5-v0',
render='human',
max_steps=1000)
model = DoubleDuelingDQN(data)
learn = AgentLearner(data, model)
learn.fit(5)
def test_basic_dqn_model_maze():
data = MDPDataBunch.from_env('maze-random-5x5-v0',
render='human',
max_steps=200)
model = DQN(data)
learn = AgentLearner(data, model)
learn.fit(5)
def test_fixed_target_dqn_model_maze():
print('\n')
data = MDPDataBunch.from_env('maze-random-5x5-v0',
render='human',
max_steps=1000)
model = FixedTargetDQN(data)
learn = AgentLearner(data, model)
learn.fit(5)
def test_interpretation_plot_q_ddpg_returns():
data = MDPDataBunch.from_env('Pendulum-v0', render='human')
# data = MDPDataBunch.from_env('MountainCarContinuous-v0', render='human')
model = DDPG(data, batch=8)
learn = AgentLearner(data, model)
learn.fit(5)
interp = AgentInterpretationAlpha(learn)
interp.plot_heatmapped_episode(2)
def test_epsilon():
data = MDPDataBunch.from_env('maze-random-5x5-v0',
render='human',
max_steps=100,
add_valid=False)
model = FixedTargetDQN(data,
batch_size=64,
max_episodes=100,
copy_over_frequency=4)
learn = AgentLearner(data, model)
learn.fit(20)
def test_ddpg():
data = MDPDataBunch.from_env('Pendulum-v0', render='human')
# data = MDPDataBunch.from_env('MountainCarContinuous-v0', render='human')
model = DDPG(data, batch=8)
learn = AgentLearner(data, model)
learn.fit(450)
def test_fit_function_ddpg():
data = MDPDataBunch.from_env('Pendulum-v0', render='human', max_steps=1000)
model = DDPG(data, memory=PriorityExperienceReplay(1000))
learn = AgentLearner(data, model)
learn.fit(5)