def solve(params, cis):
# python has dynamic typing, the line below can help IDEs with autocompletion
assert isinstance(cis, ChallengeInterfaceSolution)
# after this cis. will provide you with some autocompletion in some IDEs (e.g.: pycharm)
cis.info('Creating model.')
# you can have logging capabilties through the solution interface (cis).
# the info you log can be retrieved from your submission files.
# BEGIN SUBMISSION
# if you have a model class with a predict function this are likely the only lines you will need to modifiy
from model import TensorflowModel
# define observation and output shapes
model = TensorflowModel(
observation_shape=(1, ) +
expect_shape, # this is the shape of the image we get.
action_shape=(1, 1), # we need to output v, omega.
graph_location='tf_models/'
) # this is the folder where our models are stored.
# END SUBMISSION
try:
# We get environment from the Evaluation Engine
cis.info('Making environment')
env = gym.make(params['env'])
# Then we make sure we have a connection with the environment and it is ready to go
cis.info('Reset environment')
observation = env.reset()
check_valid_observations(observation)
cis.info('Obtained first observations.')
# While there are no signal of completion (simulation done)
# we run the predictions for a number of episodes, don't worry, we have the control on this part
while True:
# we passe the observation to our model, and we get an action in return
action = model.predict(observation)
# we tell the environment to perform this action and we get some info back in OpenAI Gym style
observation, reward, done, info = env.step(action)
check_valid_observations(observation)
# here you may want to compute some stats, like how much reward are you getting
# notice, this reward may no be associated with the challenge score.
# it is important to check for this flag, the Evalution Engine will let us know when should we finish
# if we are not careful with this the Evaluation Engine will kill our container and we will get no score
# from this submission
if 'simulation_done' in info:
cis.info('Received simulation_done.')
break
if done:
cis.info('End of episode')
env.reset()
finally:
cis.info('Releasing CPU/GPU resources.')
# release CPU/GPU resources, let's be friendly with other users that may need them
model.close()
cis.info("Graceful exit of solve().")
ACTIONS_SHAPE = (None, 1)
SEED = 1234
STORAGE_LOCATION = "trained_models/behavioral_cloning"
reader = Reader('train.log')
observations, actions = reader.read()
actions = np.array(actions)
actions = actions[:, [1]]
print(actions.shape)
observations = np.array(observations)
model = TensorflowModel(
observation_shape=OBSERVATIONS_SHAPE, # from the logs we've got
action_shape=ACTIONS_SHAPE, # same
graph_location=
STORAGE_LOCATION, # where do we want to store our trained models
seed=SEED # to seed all random operations in the model (e.g., dropout)
)
min_loss = 1000
# we trained for EPOCHS epochs
epochs_bar = tqdm(range(EPOCHS))
for i in epochs_bar:
# we defined the batch size, this can be adjusted according to your computing resources...
loss = None
for batch in range(0, len(observations), BATCH_SIZE):
loss = model.train(observations=observations[batch:batch + BATCH_SIZE],
actions=actions[batch:batch + BATCH_SIZE])
STEPS = 500
env = DuckietownEnv(
map_name='loop_empty',
max_steps=EPISODES * STEPS,
domain_rand=False,
camera_width=640,
camera_height=480,
accept_start_angle_deg=4, # start close to straight
full_transparency=True,
)
model = TensorflowModel(
observation_shape=OBSERVATIONS_SHAPE, # from the logs we've got
action_shape=ACTIONS_SHAPE, # same
graph_location=
STORAGE_LOCATION, # where do we want to store our trained models
seed=SEED # to seed all random operations in the model (e.g., dropout)
)
observation = env.reset()
# we can use the gym reward to get an idea of the performance of our model
cumulative_reward = 0.0
for episode in range(0, EPISODES):
for steps in range(0, STEPS):
action = model.predict(observation)
#print('Action', action)
#action = np.abs(action)
observation, reward, done, info = env.step(action)
def export(self):
TensorflowModel.export(self, "synth.pb")
def __init__(self, num_nodes):
self.num_nodes = num_nodes
TensorflowModel.__init__(self, "synth")
def __init__(self, num_nodes):
self.num_nodes = num_nodes
TensorflowModel.__init__(self, "resonate")
def export(self):
TensorflowModel.export(self, "resonate.pb")