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

normalize reward; square reward

This commit is contained in:
yangdsh 2019-10-14 17:24:17 +00:00 committed by Dana Van Aken
parent 090387a176
commit a3833d83b1
1 changed files with 8 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
server/website/website/tasks/async_tasks.py
View File

@ -235,8 +235,12 @@ def train_ddpg(result_id):
# Extract data from result # Extract data from result
result = Result.objects.filter(pk=result_id) result = Result.objects.filter(pk=result_id)
base_result_id = session_results[0].pk
base_result = Result.objects.filter(pk=base_result_id)
agg_data = DataUtil.aggregate_data(result) agg_data = DataUtil.aggregate_data(result)
metric_data = agg_data['y_matrix'].flatten() metric_data = agg_data['y_matrix'].flatten()
base_metric_data = (DataUtil.aggregate_data(base_result))['y_matrix'].flatten()
metric_scalar = MinMaxScaler().fit(metric_data.reshape(1, -1)) metric_scalar = MinMaxScaler().fit(metric_data.reshape(1, -1))
normalized_metric_data = metric_scalar.transform(metric_data.reshape(1, -1))[0] normalized_metric_data = metric_scalar.transform(metric_data.reshape(1, -1))[0]
@ -262,14 +266,16 @@ def train_ddpg(result_id):
'metrics (target_obj={})').format(len(target_obj_idx), 'metrics (target_obj={})').format(len(target_obj_idx),
target_objective)) target_objective))
objective = metric_data[target_obj_idx] objective = metric_data[target_obj_idx]
base_objective = base_metric_data[target_obj_idx]
metric_meta = MetricCatalog.objects.get_metric_meta(result.session.dbms, metric_meta = MetricCatalog.objects.get_metric_meta(result.session.dbms,
result.session.target_objective) result.session.target_objective)
# Calculate the reward # Calculate the reward
objective = objective / base_objective
if metric_meta[target_objective].improvement == '(less is better)': if metric_meta[target_objective].improvement == '(less is better)':
reward = -objective reward = -objective * objective
else: else:
reward = objective reward = objective * objective
LOG.info('reward: %f', reward) LOG.info('reward: %f', reward)
# Update ddpg # Update ddpg