emmi.datasets.drivaerml¶

Classes¶

`DrivAerMLStats`
`DrivAerMLDefaultSplitIDs`
`DrivAerMLDataset`	Dataset implementation for DrivAerML that supports both local and AISTORE storage.

Module Contents¶

class emmi.datasets.drivaerml.DrivAerMLStats¶

raw_pos_min: tuple[float]¶

raw_pos_max: tuple[float] = (80.0,)¶

surface_pressure_mean: tuple[float]¶

surface_pressure_std: tuple[float] = (269.345,)¶

surface_wallshearstress_mean: tuple[float, float, float]¶

surface_wallshearstress_std: tuple[float, float, float] = (2.0767, 1.35628, 1.11426)¶

volume_totalpcoeff_mean: tuple[float] = (0.171387,)¶

volume_totalpcoeff_std: tuple[float] = (0.500826,)¶

volume_velocity_mean: tuple[float, float, float]¶

volume_velocity_std: tuple[float, float, float] = (16.4115, 8.63614, 6.64996)¶

volume_vorticity_mean: tuple[float, float, float] = (3.07268, 988.601, 3.49965)¶

volume_vorticity_std: tuple[float, float, float] = (126922.0, 97991.9, 117787.0)¶

volume_vorticity_logscale_mean: tuple[float, float, float]¶

volume_vorticity_logscale_std: tuple[float, float, float] = (5.45681, 5.77081, 5.46175)¶

volume_vorticity_magnitude_mean: float = 3767.93¶

volume_vorticity_magnitude_std: float = 198927.0¶

raw_pos_z_min: float = -0.31758¶

class emmi.datasets.drivaerml.DrivAerMLDefaultSplitIDs¶

train¶

val¶

test¶

hidden_test¶

static create_split(seed=42)¶: A helper function to create a random split of the dataset. The default indices were created with seed=42.

class emmi.datasets.drivaerml.DrivAerMLDataset(split, root=None, ais_bucket_name=None, ais_endpoint='http://compute-permanent-node-179:8080', **kwargs)¶

Bases: ksuit.data.Dataset

Dataset implementation for DrivAerML that supports both local and AISTORE storage.

Currently, supports the following properties: - position of car surface points and corresponding values for pressure and wall shear stress. - position of the volumetric cells with corresponding values for pressure, velocity and vorticity.

Parameters:

split (str) – Which split to use. “debug” will use all available “run_X” folders. “train”, “valid”, “test” will load the splits as specified in the “train_val_test_splits/{split}_design_ids.txt” file.
root (str | None) – path to the processed dataset, e.g. /nfs-gpu/simulation/data/drivaerml_processed/subsampled_10x.
ais_bucket_name (str | None) – The name of the ais bucket on oracle object storage, e.g. ‘emmi-drivaerml-subsampled-10x’.
ais_endpoint (str) – The endpoint (host) running the AISTORE instance.

Args: dataset_config_provider: Optional provider for dataset configuration. dataset_normalizers: A dictionary that contains normalization ComposePreProcess(ers) for each data type. The key for each normalizer can be used in the with_normalizers decorator.

is_using_ais¶

split¶

property default_split_ids: dict[str, list[int]]¶

Property to get the sorted split IDs for train, validation, and test sets, based on DEFAULT_SPLIT_*_IDS.

Return type:: dict[str, list[int]]

static get_normalization_stats()¶

Returns the normalization stats for the dataset. Exists for backwards compatibility with downstream repo, where collators use this.

Return type:: DrivAerMLStats

getitem_surface_position_vtp(idx)¶

Retrieves surface positions (num_surface_points, 3)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_surface_position_stl(idx)¶

Retrieves surface positions (num_surface_points, 3)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_surface_position_stl_resampled100k(idx)¶

Retrieves surface positions (num_surface_points, 3)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_surface_pressure(idx)¶

Retrieves surface pressures (num_surface_points, 1)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_surface_wallshearstress(idx)¶

Retrieves surface wallshearstress (num_surface_points, 3)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_volume_position(idx)¶

Retrieves volume position (num_volume_points, 3)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_volume_totalpcoeff(idx)¶

Retrieves volume pressures (num_volume_points, 1)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_volume_velocity(idx)¶

Retrieves volume velocity (num_volume_points, 3)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_volume_vorticity(idx)¶

Retrieves volume vorticity (num_volume_points, 3)

Parameters:: idx (int)
Return type:: torch.Tensor

getitem_volume_distance_to_surface(idx)¶

Retrieves distance to surface of volume points (num_volume_points, 1)

Parameters:: idx (int)
Return type:: torch.Tensor