Description
Zarr has a native type; HDF5 does not, so scientific and robotics datasets (EgoDex, many LeRobot-style exports, most neuro/sensor corpora) are read through h5py inside a generator, which forfeits lazy column reads and pushdown. Daft ships Hdf5File with .attrs() and .to_tempfile(), which is what their EgoDex example uses.
Description
Zarr has a native type; HDF5 does not, so scientific and robotics datasets (EgoDex, many LeRobot-style exports, most neuro/sensor corpora) are read through
h5pyinside a generator, which forfeits lazy column reads and pushdown. Daft shipsHdf5Filewith.attrs()and.to_tempfile(), which is what their EgoDex example uses.