/ kernel_resolution
representation = KernelizediFDD(domain,
sparsify=sparsify,
kernel=linf_triangle_kernel,
kernel_args=[kernel_width],
active_threshold=active_threshold,
discover_threshold=discover_threshold,
normalization=True,
max_active_base_feat=10,
max_base_feat_sim=max_base_feat_sim)
policy = eGreedy(representation, epsilon=0.1)
# lambda_=.0, learn_rate_decay_mode="boyan", boyan_N0=boyan_N0)
opt["agent"] = Q_Learning(policy,
representation,
discount_factor=domain.discount_factor,
lambda_=lambda_,
initial_learn_rate=initial_learn_rate,
learn_rate_decay_mode="boyan",
boyan_N0=boyan_N0)
experiment = Experiment(**opt)
return experiment
if __name__ == '__main__':
from rlpy.Tools.run import run_profiled
run_profiled(make_experiment)
#experiment = make_experiment(1)
# experiment.run(visualize_learning=True)
# experiment.plot()
# experiment.save()
def make_experiment(
exp_id=1, path="./Results/Temp/{domain}/{agent}/{representation}/",
lambda_=0.,
boyan_N0=3571.6541,
initial_learn_rate=0.62267772):
opt = {}
opt["path"] = path
opt["exp_id"] = exp_id
opt["max_steps"] = 500000
opt["num_policy_checks"] = 30
opt["checks_per_policy"] = 10
domain = PST(NUM_UAV=4)
opt["domain"] = domain
representation = IndependentDiscretization(domain)
policy = eGreedy(representation, epsilon=0.1)
opt["agent"] = Q_Learning(policy, representation,
discount_factor=domain.discount_factor,
lambda_=lambda_, initial_learn_rate=initial_learn_rate,
learn_rate_decay_mode="boyan", boyan_N0=boyan_N0)
experiment = Experiment(**opt)
return experiment
if __name__ == '__main__':
from rlpy.Tools.run import run_profiled
run_profiled(make_experiment)
#experiment = make_experiment(1)
# experiment.run()
# experiment.plot()
# experiment.save()
opt = {}
opt["exp_id"] = exp_id
opt["path"] = path
# Domain:
maze = os.path.join(GridWorld.default_map_dir, '4x5.txt')
domain = GridWorld(maze, noise=0.3)
opt["domain"] = domain
# Representation
representation = Tabular(domain, discretization=20)
# Policy
policy = eGreedy(representation, epsilon=0.2)
# Agent
opt["agent"] = Q_Learning(representation=representation, policy=policy,
discount_factor=domain.discount_factor,
initial_learn_rate=0.1,
learn_rate_decay_mode="boyan", boyan_N0=100,
lambda_=0.)
opt["checks_per_policy"] = 100
opt["max_steps"] = 2000
opt["num_policy_checks"] = 10
experiment = Experiment(**opt)
return experiment
if __name__ == '__main__':
from rlpy.Tools.run import run_profiled
run_profiled(make_experiment, '.', 'gridworld.pdf')
domain = GridWorld(maze, noise=0.3)
opt["domain"] = domain
# Representation
representation = Tabular(domain, discretization=20)
# Policy
policy = eGreedy(representation, epsilon=0.2)
# Agent
opt["agent"] = Q_Learning(
representation=representation,
policy=policy,
discount_factor=domain.discount_factor,
initial_learn_rate=0.1,
learn_rate_decay_mode="boyan",
boyan_N0=100,
lambda_=0.0,
)
opt["checks_per_policy"] = 100
opt["max_steps"] = 2000
opt["num_policy_checks"] = 10
experiment = Experiment(**opt)
return experiment
if __name__ == "__main__":
from rlpy.Tools.run import run_profiled
run_profiled(make_experiment, ".", "gridworld.pdf")
