public
/
ottertune
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

modify the hyper parameter and reward of ddpg

This commit is contained in:
yangdsh 2019-10-14 02:58:50 +00:00 committed by Dana Van Aken
parent a5f393cdc1
commit f071a0e62c
2 changed files with 19 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
server/website/website/settings/constants.py
View File

@ -77,13 +77,7 @@ DNN_DEBUG_INTERVAL = 100
DDPG_BATCH_SIZE = 32
# Learning rate of actor network
ACTOR_LEARNING_RATE = 0.001
ACTOR_LEARNING_RATE = 0.01
# Learning rate of critic network
CRITIC_LEARNING_RATE = 0.001
# The impact of future reward on the decision
GAMMA = 0.9
# The changing rate of the target network
TAU = 0.002
CRITIC_LEARNING_RATE = 0.01

49
server/website/website/tasks/async_tasks.py
View File

@ -30,7 +30,7 @@ from website.settings import (DEFAULT_LENGTH_SCALE, DEFAULT_MAGNITUDE,
DEFAULT_EPSILON, MAX_ITER, GPR_EPS,
DEFAULT_SIGMA_MULTIPLIER, DEFAULT_MU_MULTIPLIER,
DDPG_BATCH_SIZE, ACTOR_LEARNING_RATE,
CRITIC_LEARNING_RATE, GAMMA, TAU,
CRITIC_LEARNING_RATE,
DNN_TRAIN_ITER, DNN_EXPLORE, DNN_EXPLORE_ITER,
DNN_NOISE_SCALE_BEGIN, DNN_NOISE_SCALE_END,
DNN_DEBUG, DNN_DEBUG_INTERVAL)
@ -232,22 +232,18 @@ def train_ddpg(result_id):
if len(session_results) == 0:
LOG.info('No previous result. Abort.')
return result_info
prev_result_id = session_results[len(session_results) - 1].pk
base_result_id = session_results[0].pk
prev_result = Result.objects.filter(pk=prev_result_id)
base_result = Result.objects.filter(pk=base_result_id)
# Extract data from result
result = Result.objects.filter(pk=result_id)
agg_data = DataUtil.aggregate_data(result)
metric_data = agg_data['y_matrix'].flatten()
prev_metric_data = (DataUtil.aggregate_data(prev_result))['y_matrix'].flatten()
base_metric_data = (DataUtil.aggregate_data(base_result))['y_matrix'].flatten()
metric_scalar = MinMaxScaler().fit(metric_data.reshape(1, -1))
normalized_metric_data = metric_scalar.transform(metric_data.reshape(1, -1))[0]
# Clean knob data
cleaned_agg_data = clean_knob_data(agg_data['X_matrix'], agg_data['X_columnlabels'], session)
knob_data = np.array(cleaned_agg_data[0])
knob_labels = np.array(cleaned_agg_data[1])
cleaned_knob_data = clean_knob_data(agg_data['X_matrix'], agg_data['X_columnlabels'], session)
knob_data = np.array(cleaned_knob_data[0])
knob_labels = np.array(cleaned_knob_data[1])
knob_bounds = np.vstack(DataUtil.get_knob_bounds(knob_labels.flatten(), session))
knob_data = MinMaxScaler().fit(knob_bounds).transform(knob_data)[0]
knob_num = len(knob_data)
@ -266,34 +262,24 @@ def train_ddpg(result_id):
'metrics (target_obj={})').format(len(target_obj_idx),
target_objective))
objective = metric_data[target_obj_idx]
prev_objective = prev_metric_data[target_obj_idx]
base_objective = base_metric_data[target_obj_idx]
metric_meta = MetricCatalog.objects.get_metric_meta(result.session.dbms,
result.session.target_objective)
# Calculate the reward
reward = 0
if metric_meta[target_objective].improvement == '(less is better)':
if objective - base_objective <= 0:
reward = -(np.square(objective / base_objective) - 1) * objective / prev_objective
reward = -objective
else:
reward = (np.square((2 * base_objective - objective) / base_objective) - 1)\
* (2 * prev_objective - objective) / prev_objective
else:
if objective - base_objective > 0:
reward = (np.square(objective / base_objective) - 1) * objective / prev_objective
else:
reward = -(np.square((2 * base_objective - objective) / base_objective) - 1)\
* (2 * prev_objective - objective) / prev_objective
reward = objective
LOG.info('reward: %f', reward)
# Update ddpg
ddpg = DDPG(n_actions=knob_num, n_states=metric_num, alr=ACTOR_LEARNING_RATE,
clr=CRITIC_LEARNING_RATE, gamma=GAMMA, batch_size=DDPG_BATCH_SIZE, tau=TAU)
clr=CRITIC_LEARNING_RATE, gamma=0.0, batch_size=DDPG_BATCH_SIZE, tau=0.0)
if session.ddpg_actor_model and session.ddpg_critic_model:
ddpg.set_model(session.ddpg_actor_model, session.ddpg_critic_model)
if session.ddpg_reply_memory:
ddpg.replay_memory.set(session.ddpg_reply_memory)
ddpg.add_sample(prev_metric_data, knob_data, reward, metric_data, False)
ddpg.add_sample(normalized_metric_data, knob_data, reward, normalized_metric_data, False)
if len(ddpg.replay_memory) > 32:
ddpg.update()
session.ddpg_actor_model, session.ddpg_critic_model = ddpg.get_model()
@ -310,20 +296,21 @@ def configuration_recommendation_ddpg(result_info): # pylint: disable=invalid-n
session = Result.objects.get(pk=result_id).session
agg_data = DataUtil.aggregate_data(result)
metric_data = agg_data['y_matrix'].flatten()
cleaned_agg_data = clean_knob_data(agg_data['X_matrix'], agg_data['X_columnlabels'],
metric_scalar = MinMaxScaler().fit(metric_data.reshape(1, -1))
normalized_metric_data = metric_scalar.transform(metric_data.reshape(1, -1))[0]
cleaned_knob_data = clean_knob_data(agg_data['X_matrix'], agg_data['X_columnlabels'],
session)
knob_labels = np.array(cleaned_agg_data[1]).flatten()
knob_labels = np.array(cleaned_knob_data[1]).flatten()
knob_num = len(knob_labels)
metric_num = len(metric_data)
ddpg = DDPG(n_actions=knob_num, n_states=metric_num, alr=ACTOR_LEARNING_RATE,
clr=CRITIC_LEARNING_RATE, gamma=GAMMA, batch_size=DDPG_BATCH_SIZE, tau=TAU)
clr=CRITIC_LEARNING_RATE, gamma=0.0, batch_size=DDPG_BATCH_SIZE, tau=0.0)
if session.ddpg_actor_model is not None and session.ddpg_critic_model is not None:
ddpg.set_model(session.ddpg_actor_model, session.ddpg_critic_model)
if session.ddpg_reply_memory is not None:
ddpg.replay_memory.set(session.ddpg_reply_memory)
knob_data = ddpg.choose_action(metric_data)
LOG.info('recommended knob: %s', knob_data)
knob_data = ddpg.choose_action(normalized_metric_data)
knob_bounds = np.vstack(DataUtil.get_knob_bounds(knob_labels, session))
knob_data = MinMaxScaler().fit(knob_bounds).inverse_transform(knob_data.reshape(1, -1))[0]
@ -333,8 +320,6 @@ def configuration_recommendation_ddpg(result_info): # pylint: disable=invalid-n
conf_map_res['result_id'] = result_id
conf_map_res['recommendation'] = conf_map
conf_map_res['info'] = 'INFO: ddpg'
for k in knob_labels:
LOG.info('%s: %f', k, conf_map[k])
return conf_map_res