Skip to content

API Reference

The RDDAC package exposes a small set of public functions, all importable from the top level rddac module. The surface mirrors the ddacs package one to one, so DDACS code ports by swapping the import.

Modules

  • Croissant: load and add_view. Loads the Croissant manifest and adds custom RecordSets.
  • HDF5: open_h5, inspect_h5. Reads a single experiment by id and prints the HDF5 hierarchy.
  • Streaming: iter_view, export_to_numpy, export_to_numpy_per_sim, load_export. Torch-free iteration and numpy materialisation.
  • PyTorch: RDDACDataset. Streaming IterableDataset over a Croissant view. Requires the [torch] extra.
  • Visualization: plot_scan, scan_to_pointcloud, plot_point_cloud, plot_force, plot_traverse. Matplotlib plotting helpers.
  • Preprocessing: the rddac preprocess CLI and the internal rddac._preprocess stages (oil, force, sheet, pointcloud) plus the processing-figure helpers. The pointcloud stage requires the [preprocessing] extra.

Quick import

import rddac

# Croissant entry points
ds = rddac.load(data_dir="./data")
rddac.add_view(ds, "my-view", fields={"force": "force_data"})

# Single-experiment HDF5 access
with rddac.open_h5(0) as f:
    rddac.inspect_h5(f)

# Streaming (no PyTorch required)
for rec in rddac.streaming.iter_view("force-curve", data_dir="./data", dataset=ds):
    ...

# Visualization
ax, cbar = rddac.plot_scan(z_buffer)

# PyTorch (requires `pip install rddac[torch]`)
from rddac.pytorch import RDDACDataset
loader = RDDACDataset(view="force-curve")

Preprocessing has no Python entry point on purpose: rddac preprocess is the supported interface (see Preprocessing); the same views then stream the processed layer via data_dir=.