def run_experiment(Lambda,
alpha,
twe,
trunc_normal,
subspaces,
num_runs,
num_episodes=20000,
num_procs=None,
name=""):
returns = np.empty((num_runs, num_episodes), dtype=np.float64)
results.append(returns)
for i in xrange(num_runs):
print name
agent = PolicyGradientAgent(simulator,
Lambda=Lambda,
alpha_u=alpha,
alpha_v=alpha,
tile_weight_exponent=twe,
trunc_normal=trunc_normal,
subspaces=subspaces)
agent.persist_state()
framework = Framework(simulator, agent, num_episodes=num_episodes)
framework.train(num_procs=num_procs)
returns[i] = framework.returns
random = np.random.randint(sys.maxsize)
directory = 'data/%s/' % (name)
filename = directory + ('%d.npy' % (random))
try:
os.makedirs(directory)
except OSError:
pass
np.save(filename, returns)
return returns
def run_experiment(Lambda, alpha, twe, trunc_normal, subspaces, num_runs,num_episodes=20000, num_procs=None,name=""):
returns = np.empty((num_runs, num_episodes), dtype=np.float64)
results.append(returns)
for i in xrange(num_runs):
print(name)
agent = PolicyGradientAgent (simulator,
Lambda=Lambda, alpha_u=alpha, alpha_v=alpha,
tile_weight_exponent=twe,
trunc_normal=trunc_normal,
subspaces=subspaces)
agent.persist_state()
framework = Framework(simulator, agent, num_episodes=num_episodes)
framework.train(num_procs=num_procs)
returns[i] = framework.returns
random = np.random.randint(sys.maxsize)
directory = 'data/%s/' % (name)
filename = directory + ('%d.npy' % (random))
try:
os.makedirs(directory)
except OSError:
pass
np.save (filename, returns)
return returns
