buildings_bench.utils

Common Utilities

buildings_bench.utils

get_puma_county_lookup_table(metadata_dir)

Build a puma-county lookup table.

The weather files are organized by U.S. county. We need to map counties to PUMAs and ensure we drop counties without weather data (only done when using weather).

Parameters:

Name	Type	Description	Default
metadata_dir	Path	Path to the metadata folder of BuildingsBench	required

Returns:

Type	Description
DataFrame	pd.DataFrame: the puma-county lookup table

load_model_checkpoint(path, model, optimizer, scheduler, local_rank)

Load model checkpoint.

save_model_checkpoint(model, optimizer, scheduler, step, path)

Save model checkpoint.

set_seed(seed=42)

Set random seed for reproducibility.

time_features_to_datetime(time_features, year)

Convert time features to datetime objects.

Parameters:

Name	Type	Description	Default
time_features	ndarray	Array of time features. [:,0] is day of year, [:,1] is day of week, [:,2] is hour of day.	required
year	int	Year to use for datetime objects.	required

Returns:

Type	Description
array	np.array: Array of datetime objects.

worker_init_fn_eulp(worker_id)

Set random seed for each worker and init file pointer for Buildings-900K dataset workers.

Parameters:

Name	Type	Description	Default
worker_id	int	worker id	required