Skip to content

Optimize data processing and improve logging for autoheuristic training #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ShashankFC wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Optimize data processing and improve logging for autoheuristic training #2

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 9 additions & 5 deletions torchgen/_autoheuristic/train_decision.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,8 +219,8 @@ def eval(name, df):
print(
f"new best model with {num_correct} correct and {num_wrong} wrong"
)
best_model = model
best_model_num_correct = num_correct
best_model = model
best_model_safe_proba = safe_proba
best_model_unsafe_leaves = unsafe_leaves

Expand Down Expand Up @@ -403,7 +403,7 @@ def calculate_stats(group):
if relative_std > 5:
times = group["feedback"].tolist()
times_str = ", ".join([f"{t:.3f}" for t in sorted(times)])
log.debug("High relative std: %f. times=%s", relative_std, times_str)
log.error("High relative std: %f. times=%s. group=%s", relative_std, times_str, group.to_dict())
return pd.Series(
{
"count": count,
Expand Down Expand Up @@ -439,12 +439,15 @@ def get_winner_and_speedup(group):
winner = sorted_group.iloc[0]["choice"]
winning_time = sorted_group.iloc[0]["median_execution_time"]
second_best_time = sorted_group.iloc[1]["median_execution_time"]
speedup = second_best_time / winning_time
speedup = second_best_time / winning_time if winning_time > 0 else 1.0
unique_choices = group["choice"].unique()

choice2time = {}
for row in group.itertuples():
choice2time[row.choice] = row.median_execution_time
for other_row in group.itertuples():
if row.choice == other_row.choice:
choice2time[row.choice] = row.median_execution_time
break

assert len(unique_choices) == len(group), (
f"len(unique_choices) != len(group): {len(unique_choices)} != {len(group)}"
Expand Down Expand Up @@ -482,8 +485,9 @@ def add_near_best_configs(df):
breakpoint()
new_row["actual_winner"] = row["winner"]
new_row["winner"] = key
new_rows.append(new_row)
if relative_performance >= 0.98:
new_rows.append(new_row)
pass

return pd.DataFrame(new_rows).reset_index(drop=True)

Expand Down