Skip to content

[1438] Adding padding to end_date to avoid duplicate samples #1749

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
enssow wants to merge 1 commit into
base: develop
Choose a base branch
from
Open

[1438] Adding padding to end_date to avoid duplicate samples #1749

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
30 changes: 22 additions & 8 deletions src/weathergen/datasets/multi_stream_data_sampler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,14 +92,6 @@ def __init__(
self.batch_size = get_batch_size_from_config(mode_cfg)
self.len_timedelta: np.timedelta64 = mode_cfg.time_window_len
self.step_timedelta: np.timedelta64 = mode_cfg.time_window_step
self.time_window_handler = TimeWindowHandler(
mode_cfg.start_date, mode_cfg.end_date, self.len_timedelta, self.step_timedelta
)
if is_root():
logger.info(self.time_window_handler)

index_range = self.time_window_handler.get_index_range()
perms_len = int(index_range.end - index_range.start)

# Handle forecast_delta_hrs which might be int (hours) or string (timedelta)
self.forecast_cfg = mode_cfg.get("forecast", {})
Expand All @@ -125,6 +117,28 @@ def __init__(

fsm = self.list_num_forecast_steps[0]
forecast_len = (self.time_step * (fsm + 1)) // self.step_timedelta

# Handle too short time window causing duplication of samples
available_samples = (mode_cfg.end_date - mode_cfg.start_date) // self.time_step
if self.samples_per_mini_epoch >= available_samples:
# padding to widen time window
samples_diff = (
self.samples_per_mini_epoch - available_samples + forecast_len + 1
) # extra step to be safe
new_end_date = mode_cfg.end_date + (samples_diff * self.time_step)
logger.warning(f"Using adjusted end date {new_end_date} instead of {mode_cfg.end_date}")
self.time_window_handler = TimeWindowHandler(
mode_cfg.start_date, new_end_date, self.len_timedelta, self.step_timedelta
)
else:
self.time_window_handler = TimeWindowHandler(
mode_cfg.start_date, mode_cfg.end_date, self.len_timedelta, self.step_timedelta
)
if is_root():
logger.info(self.time_window_handler)

index_range = self.time_window_handler.get_index_range()
perms_len = int(index_range.end - index_range.start)
perms_len = perms_len - (forecast_len + self.output_offset)

self.repeat_data = cf.data_loading.get("repeat_data_in_mini_epoch", False)
Expand Down
Loading