def read_c4_dataset_as_c8(ds: h5py.Dataset, key=np.s_[...]):
"""
Read a complex float16 HDF5 dataset as a numpy.complex64 array.
Avoids h5py/numpy dtype bugs and uses numpy float16 -> float32 conversions
which are about 10x faster than HDF5 ones.
"""
# This context manager avoids h5py exception:
# TypeError: data type '<c4' not understood
with ds.astype(complex32):
z = ds[key]
# Define a similar datatype for complex64 to be sure we cast safely.
complex64 = np.dtype([("r", np.float32), ("i", np.float32)])
# Cast safely and then view as native complex64 numpy dtype.
return z.astype(complex64).view(np.complex64)
def prop_to_dataframe(dset: h5py.Dataset,
dtype: DTypeLike = None) -> pd.DataFrame:
"""Convert the passed property Dataset into a DataFrame.
Examples
--------
.. testsetup:: python
>>> from dataCAT.testing_utils import HDF5_READ as hdf5_file
.. code:: python
>>> import h5py
>>> from dataCAT import prop_to_dataframe
>>> hdf5_file = str(...) # doctest: +SKIP
>>> with h5py.File(hdf5_file, 'r') as f:
... dset = f['ligand/properties/E_solv']
... df = prop_to_dataframe(dset)
... print(df) # doctest: +NORMALIZE_WHITESPACE
E_solv_names water methanol ethanol
ligand ligand anchor
O=C=O O1 -0.918837 -0.151129 -0.177396
O3 -0.221182 -0.261591 -0.712906
CCCO O4 -0.314799 -0.784353 -0.190898
Parameters
----------
dset : :class:`h5py.Dataset`
The property-containing Dataset of interest.
dtype : dtype-like, optional
The data type of the to-be returned DataFrame.
Use :data:`None` to default to the data type of **dset**.
Returns
-------
:class:`pandas.DataFrame`
A DataFrame constructed from the passed **dset**.
""" # noqa: E501
# Construct the index
dim0 = dset.dims[0]
scale0 = dim0[0]
index = index_to_pandas(scale0)
# Construct the columns
if dset.ndim == 1:
full_name = dset.name
name = full_name.rsplit('/', 1)[-1]
columns = pd.Index([name])
else:
dim1 = dset.dims[1]
scale1 = dim1[0]
columns = pd.Index(scale1[:].astype(str), name=dim1.label)
# Create and return the dataframe
if dtype is None:
return pd.DataFrame(dset[:], index=index, columns=columns)
# If possible, let h5py handle the datatype conversion
# This will often fail when dset.dtype consists of variable-length bytes-strings
try:
with dset.astype(dtype):
return pd.DataFrame(dset[:], index=index, columns=columns)
except (ValueError, TypeError):
return pd.DataFrame(dset[:].astype(dtype),
index=index,
columns=columns)
