PhysArena/PDEBench_2D_SWE

legal:
  owner: Takamoto, M et al. (https://darus.uni-stuttgart.de/dataset.xhtml?persistentId=doi:10.18419/darus-2986)
  license: cc-by-4.0
data_production:
  physics: Shallow Water Equations
  type: simulation
  script: Converted to PLAID format for standardized usage; no changes to data content.
num_samples:
  train: 1000
storage_backend: hf_datasets
plaid:
  version: 0.1.12

This dataset was generated with plaid, we refer to this documentation for additional details on how to extract data from plaid_sample objects.

The simplest way to use this dataset is to first download it:

from plaid.storage import download_from_hub

repo_id = "channel/dataset"
local_folder = "downloaded_dataset"

download_from_hub(repo_id, local_folder)

Then, to iterate over the dataset and instantiate samples:

from plaid.storage import init_from_disk

local_folder = "downloaded_dataset"
split_name = "train"

datasetdict, converterdict = init_from_disk(local_folder)

dataset = datasetdict[split]
converter = converterdict[split]

for i in range(len(dataset)):
    plaid_sample = converter.to_plaid(dataset, i)

It is possible to stream the data directly:

from plaid.storage import init_streaming_from_hub

repo_id = "channel/dataset"

datasetdict, converterdict = init_streaming_from_hub(repo_id)

dataset = datasetdict[split]
converter = converterdict[split]

for sample_raw in dataset:
    plaid_sample = converter.sample_to_plaid(sample_raw)

Plaid samples' features can be retrieved like the following:

from plaid.storage import load_problem_definitions_from_disk
local_folder = "downloaded_dataset"
pb_defs = load_problem_definitions_from_disk(local_folder)

# or
from plaid.storage import load_problem_definitions_from_hub
repo_id = "channel/dataset"
pb_defs = load_problem_definitions_from_hub(repo_id)


pb_def = pb_defs[0]

plaid_sample = ... # use a method from above to instantiate a plaid sample

for t in plaid_sample.get_all_time_values():
    for path in pb_def.get_in_features_identifiers():
        plaid_sample.get_feature_by_path(path=path, time=t)
    for path in pb_def.get_out_features_identifiers():
        plaid_sample.get_feature_by_path(path=path, time=t)

For those familiar with HF's datasets library, raw data can be retrieved without using the plaid library:

from datasets import load_dataset

repo_id = "channel/dataset"

datasetdict = load_dataset(repo_id)

for split_name, dataset in datasetdict.items():
    for raw_sample in dataset:
        for feat_name in dataset.column_names:
            feature = raw_sample[feat_name]

Notice that raw data refers to the variable features only, with a specific encoding for time variable features.

Dataset Sources

• Papers:
  • arxiv