training_steps = 5000
eval_steps = 2000
load_period = 250
epsilon = 0.5
MIN_VMS = 1
MAX_VMS = 20
split_crit = ANY_POINT
cons_trans = True
CONF_FILE = TIRAMOLA_DIR + "read_load_5k/any.json"
##############################################################
PRINT = len(sys.argv) > 1 and sys.argv[1] == "-p"
dm = DecisionMaking.DecisionMaker(TIRAMOLA_DIR + "decisionMaking.json",
'training.data')
scenario = ReadLoadScenario(training_steps, load_period, 10, MIN_VMS, MAX_VMS)
dm.set_state(scenario.get_current_measurements())
model = dm.get_model()
model.set_splitting(split_crit, cons_trans)
total_reward = 0
if TRAIN:
for time in range(training_steps):
if random.uniform(0, 1) < epsilon:
action = random.choice(model.get_legal_actions())
else:
action = model.suggest_action()
reward = scenario.execute_action(action)
meas = scenario.get_current_measurements()
"start", "start_chain", "half", "half_chain", "end", "always"
]:
print "Unknown split strategy!"
exit()
if "start" in split_strategy:
split_step = 0
elif "half" in split_strategy:
split_step = training_steps // 2
else:
split_step = training_steps
PRINT = len(sys.argv) > 1 and sys.argv[1] == "-p"
if PRINT: num_tests = 1
scenario = ReadLoadScenario(training_steps)
conf = ModelConf(CONF_FILE)
assert conf.get_model_type() == MDP_DT, "Wrong model type in MDP-DT example"
total_reward_results = []
total_splits_results = []
good_splits_results = []
for i in range(num_tests):
model = MDPDTModel(conf.get_model_conf())
model.set_state(scenario.get_current_measurements())
model.set_allow_splitting(False)
model.set_splitting(split_crit, cons_trans)
total_reward = 0
CONF_FILE = TIRAMOLA_DIR + "read_load_5k/qdt.json"
##############################################################
PRINT = len(sys.argv) > 1 and sys.argv[1] == "-p"
if PRINT: num_tests = 1
conf = ModelConf(CONF_FILE)
assert conf.get_model_type() == Q_DT, "Wrong model type in QDT example"
total_reward_results = []
total_splits_results = []
good_splits_results = []
for i in range(num_tests):
scenario = ReadLoadScenario(training_steps, load_period, 10, MIN_VMS,
MAX_VMS)
model = QDTModel(conf.get_model_conf())
model.set_state(scenario.get_current_measurements())
model.set_allow_splitting(False)
total_reward = 0
for time in range(training_steps + eval_steps):
if random.uniform(0, 1) < epsilon and time < training_steps:
action = random.choice(model.get_legal_actions())
else:
action = model.suggest_action()
reward = scenario.execute_action(action)
meas = scenario.get_current_measurements()
model.update(action, meas, reward)