def createExperimentInstance():
gymRawEnv = gym.make('MountainCar-v0')
cartPositionGroup = Digitizer.buildBins(-1.2, 0.6, 16)
cartVelocityGroup = Digitizer.buildBins(-0.07, 0.07, 16)
# print("Cart position bins:", cartPositionGroup)
# print("Cart velocity bins:", cartVelocityGroup)
observationDigitizer = ArrayDigitizer(
[cartPositionGroup, cartVelocityGroup])
transformation = EnvTransformation(observationDigitizer)
task = GymTask.createTask(gymRawEnv)
env = task.env
env.setTransformation(transformation)
# env.setCumulativeRewardMode()
# create value table and initialize with ones
table = ActionValueTable(observationDigitizer.states, env.numActions)
table.initialize(0.0)
# table.initialize( np.random.rand( table.paramdim ) )
agent = createAgent(table)
experiment = Experiment(task, agent)
experiment = ProcessExperiment(experiment, ExperimentIteration())
return experiment
def createExperimentInstance():
gymRawEnv = gym.make('MountainCarContinuous-v0')
cartPositionGroup = Digitizer.buildBins(-1.2, 0.6, 16)
cartVelocityGroup = Digitizer.buildBins(-0.07, 0.07, 4)
actionDedigitizer = Digitizer.build(-1.0, 1.0, 5, True)
# print("Cart position bins:", cartPositionGroup)
# print("Cart velocity bins:", cartVelocityGroup)
# print("Cart force bins:", actionDedigitizer.bins, actionDedigitizer.possibleValues())
observationDigitizer = ArrayDigitizer(
[cartPositionGroup, cartVelocityGroup])
transformation = EnvTransformation(observationDigitizer, actionDedigitizer)
task = GymTask.createTask(gymRawEnv)
env = task.env
env.setTransformation(transformation)
# env.setCumulativeRewardMode()
# create agent with controller and learner - use SARSA(), Q() or QLambda() here
## alpha -- learning rate (preference of new information)
## gamma -- discount factor (importance of future reward)
# create value table and initialize with ones
table = ActionValueTable(observationDigitizer.states,
actionDedigitizer.states)
table.initialize(0.0)
# table.initialize( np.random.rand( table.paramdim ) )
agent = createAgent(table)
experiment = Experiment(task, agent)
experiment = ProcessExperiment(experiment, doSingleExperiment)
return experiment
## Gym observation: [position, velocity, pole angle, pole velocity]
## position: (-2.5, 2.5)
## velocity (-inf, inf)
## pole angle (-41.8, 41.8)
## pole vel: (-inf, inf)
## Reward:
## A reward of +1 is provided for every timestep that the pole remains upright.
gymRawEnv = gym.make('CartPole-v1')
# gymRawEnv = gym.make('CartPole-v0')
## env.tags['wrapper_config.TimeLimit.max_episode_steps'] = 500
cartPositionGroup = Digitizer.buildBins(
-2.4, 2.4, 6) ## terminates when outside range (-2.4, 2.4)
cartVelocityGroup = Digitizer.buildBins(-1.0, 1.0, 3)
poleAngleGroup = Digitizer.buildBins(
-12.0, 12.0, 2) ## terminates when outside range (-12, 12)
poleVelocityGroup = Digitizer.buildBins(-4.0, 4.0, 4)
print("Cart position bins:", cartPositionGroup)
print("Cart velocity bins:", cartVelocityGroup)
print("Pole angle bins:", poleAngleGroup)
print("Pole velocity bins:", poleVelocityGroup)
observationDigitizer = ArrayDigitizer(
[cartPositionGroup, cartVelocityGroup, poleAngleGroup, poleVelocityGroup])
transformation = EnvTransformation(observationDigitizer)
task = GymTask.createTask(gymRawEnv)
def test_values_noEdges_5(self):
digitizer = Digitizer.build(-1.0, 1.0, 5, False)
npt.assert_array_almost_equal(digitizer.values, [-0.6, -0.3, 0.0, 0.3, 0.6], 3)
## =============================================================================
## Gym expected action:
## 0 -- left
## 1 -- neutral
## 2 -- right
## Gym observation: [position, velocity]
## position: (-1.2, 0.6)
## velocity (-0.07, 0.07)
gymRawEnv = gym.make('MountainCar-v0')
cartPositionGroup = Digitizer.buildBins(-1.2, 0.6, 16)
cartVelocityGroup = Digitizer.buildBins(-0.07, 0.07, 16)
# print("Cart position bins:", cartPositionGroup)
# print("Cart velocity bins:", cartVelocityGroup)
observationDigitizer = ArrayDigitizer([cartPositionGroup, cartVelocityGroup])
transformation = EnvTransformation(observationDigitizer)
task = GymTask.createTask(gymRawEnv)
env = task.env
env.setTransformation(transformation)
## env.setCumulativeRewardMode()
# create value table and initialize with ones
table = ActionValueTable(observationDigitizer.states, env.numActions)
def test_value_states(self):
digitizer = Digitizer( [0.0, 2.0] )
value = digitizer.value( digitizer.states - 1 )
npt.assert_array_almost_equal(value, 2.0, 2)
def test_values_edges_5(self):
digitizer = Digitizer.build(-1.0, 1.0, 5, True)
npt.assert_array_almost_equal(digitizer.values, [-1.0, -0.5, 0.0, 0.5, 1.0], 3)
def test_value_badInput_positive(self):
digitizer = Digitizer( [0.0, 2.0] )
self.assertRaises( ValueError, digitizer.value, 3 )
def test_value_2(self):
digitizer = Digitizer( [0.0, 2.0] )
value = digitizer.value(2)
npt.assert_array_almost_equal(value, 2.0, 2)
def test_numstates(self):
digitizer = Digitizer([0.0])
states = digitizer.numstates()
self.assertEqual(states, 2)
def test_state(self):
digitizer = Digitizer( [0.0] )
indexes = digitizer.digitize( [-1.0, 1.0] )
npt.assert_equal(indexes, [0, 1])
def test_build_edge_5(self):
digitizer = Digitizer.build(0.0, 10.0, 5, True)
self.assertEqual(digitizer.states, 5)
npt.assert_array_almost_equal(digitizer.bins, [0., 3.333, 6.667, 10.], 3)
npt.assert_array_almost_equal(digitizer.values, [0., 2.5, 5., 7.5, 10.], 3)
def test_build_noEdge_5(self):
digitizer = Digitizer.build(0.0, 10.0, 5, False)
self.assertEqual(digitizer.states, 5)
npt.assert_array_almost_equal(digitizer.bins, [2., 4., 6., 8.], 3)
npt.assert_array_almost_equal(digitizer.values, [2., 3.5, 5., 6.5, 8.], 3)
def test_buildBins_bins04_edges(self):
bins = Digitizer.buildBins(0.0, 12.0, 4, True)
npt.assert_equal(bins, [0.0, 6.0, 12.0])
def test_buildBins_bins03(self):
bins = Digitizer.buildBins(0.0, 12.0, 3)
npt.assert_equal(bins, [4.0, 8.0])
def test_buildBins_bins02(self):
bins = Digitizer.buildBins(0.0, 12.0, 2)
npt.assert_equal(bins, [6.0])
