cat2import – Convert HDF5 files to Caterva2 roots

In the ecosystem where Caterva2 belongs, it is common to work with HDF5 files and other formats containing multidimensional numerical data stored as datasets arranged in arbitrary hierarchies. Caterva2 is designed to distribute such type of datasets (and some others), with similar features such as compression, chunking and arbitrary attributes, and also grouped in hierarchies.

Although Caterva2 does allow publishing an HDF5 file directly as a root (with datasets converted to Blosc2 arrays on-the-fly), it also includes cat2import, a simple tool targeted at exporting datasets in other formats into an equivalent Caterva2 root directory. Still in its early stages of development, it only supports HDF5 for the moment. To use cat2import, the tools extra needs to be installed:

python -m pip install caterva2[tools]

For the moment it has a very simple invocation syntax:

cat2import HDF5_FILE CATERVA2_ROOT

Where the HDF5 file must exist beforehand, and the Caterva2 root directory must not, as it will be created from scratch. HDF5 groups will be mapped to directories of the same name in the Caterva2 root, and datasets to Blosc2 array files with the same name, plus a .b2nd extension. There is currently no way of controlling the involved compression algorithm or parameters, nor the chunking/blocking of data, so defaults are used (this will change in the future).

Invoking cat2import --help provides more hints on which HDF5 features are supported, and how they are mapped into the Caterva2 root.

Usage example

We shall export one of the HDF5 files in the PyTables test suite: ex-noattr.h5. Download it, place it into some working directory and open a shell into that directory. Since Caterva2 tools depend on h5py, we’ll use it to look at the hierarchy in the file:

python -c 'import h5py; h5py.File("ex-noattr.h5").visit(print)'

This should show a couple of groups with datasets in them:

columns
columns/TDC
columns/name
columns/pressure
detector
detector/table

Now we shall export ex-noattr.h5 into a new Caterva2 root called ex-noattr. Just run:

cat2import ex-noattr.h5 ex-noattr

The program only outputs the following error:

ERROR:root:Failed to convert dataset to Blosc2 ND array: 'detector/table' -> ValueError('invalid shape in fixed-type tuple.')

However, it does finish successfully and it generates the requested root. Running find ex-noattr to print all files and directories under it shows something like this:

ex-noattr
ex-noattr/columns
ex-noattr/columns/TDC.b2nd
ex-noattr/columns/pressure.b2nd
ex-noattr/columns/name.b2nd
ex-noattr/detector

So the program was able to export all groups and datasets save for detector/table, hence the previous error. In general, cat2import tries to convert each group, dataset or attribute at hand, but if it fails because of some error or the conversion not being supported, it just reports the issue and continues with the next object. This behavior will be refined in the future.

Now you should be able to configure ex-noattr as the data directory for your Caterva2 publisher. See Running independent Caterva2 services for more information.