h5yaml 0.5.0

Generate and design netCDF4 files using a YAML configuration file.

H5Yaml

Description

This Python package let you design the abstract data model of your HDF5/netCDF4 files. Where the Abstract Data Model is a conceptual model of data, data types, and data organization. We choose the human-readable data serialization language YAML to define the Abstract Data Model with the following components:

• Groups which define its Hierarchical structure
• Dimensions the size(s) of the variables
• Variables the datasets to hold your data
• Attributes the meta data of the File, Groups, or Variables

From the YAML files, you can create HDF5 or netCDF4 files, which are small because the Variables may still be empty. Thus the abstract data model and storage model can be shared among your colleagues for review, or for metadata compliance checks. For example the CF conventions or the ACDD, by using:

• https://mcc.podaac.earthdatacloud.nasa.gov

And finally, you can implement the file-structure in JAVA, C++ or Fortran. But of course you can also simply generate the empty products and fill the datasets using Python.

In short, this approach has the following advantages:

• you define the layout of your HDF5/netCDF4 file using YAML which is human-readable and has intuitive syntax.
• you can reuse the YAML configuration file to to have all your product have a consistent layout.
• you can make updates by only changing the YAML configuration file.
• you can have the layout of your HDF5/netCDF4 file as a Python dictionary, thus without accessing any HDF5/netCDF4 file.

The package h5yaml provides the classes YamlToH5 and YamlToNc to generate a HDF5/netCDF4 formatted file from a Python dictionary.

1. The class YamlToH5 uses the h5py package, which is a Pythonic interface to the HDF5 binary data format. The generated HDF5 file should be compatible with the netCDF4 format. YamlToH5 is slightly faster than the netCDF4 implementation and generates smaller files.
2. The class YamlToNc uses the netCDF4 package, which provides an object-oriented python interface to the netCDF version 4 library. You should use this class when strict conformance with the netCDF4 format is required. However, package netCDF4 has some limitations, which h5py has not, for example it does not allow variable-length variables to have a compound data-type.
3. Both classes inherit the class ReadNcYaml, which read the Yaml files and combine the data in to a Python dictionary.

Installation

The package h5yaml is available from PyPI. To install it use pip:

$ pip install [--user] h5yaml

The module h5yaml requires Python3.10+ and Python modules: h5py (v3.14+), netCDF4 (v1.7+) and numpy (v2.0+).

Usage

Example 1) Use the class YamlToNc to generate a template netCDF4 file:

from importlib.resources import files

from h5yaml.yaml_to_nc import YamlToNc

yaml_list = [
   files("h5yaml.Data") / "nc_testing.yaml"),
   files("h5yaml.Data") / "h5_global_attrs.yaml",
]
# show the YAML configuration as a Python dictionary using pprint
aa = ReadNcYaml(yaml_list)
print(repr(aa))

# generate an in-memory HDF5 file
aa = YamlToNc(yaml_list)
fid = aa.diskless()

# Optional, set any unlimited dimension to a fixed length
aa.set_dims()

# write data to datasets of the file
# ...

# write netCDF4 file to disk
res.to_disk(fid, filename)

Example 2) Use the class YamlToH5 to generate a template HDF5 file:

from importlib.resources import files

from h5yaml.yaml_to_h5 import YamlToH5

yaml_list = [
   files("h5yaml.Data") / "h5_testing.yaml"),
   files("h5yaml.Data") / "h5_global_attrs.yaml",
]

# show the YAML configuration as a Python dictionary using pprint
aa = ReadNcYaml(yaml_list)
print(repr(aa))

# generate an in-memory HDF5 file
aa = YamlToH5(yaml_list)
fid = aa.diskless()

# Optional, set any unlimited dimension to a fixed length
aa.set_dims()

# write data to datasets of the file
# ...

# write HDF5 file to disk
res.to_disk(fid, filename)

Example 3) Use the class YamlToH5 to generate a template netCDF4 file on disk and add data later.

from importlib.resources import files

from netCDF4 import Dataset

from h5yaml.nc_from_yaml import NcFromYaml

res = YamlToNc(files("h5yaml.Data") / "nc_testing.yaml").create(filename)
with Dataset(filename, "r+") as fid
  # write data to variables of the file
  ...

The YAML file should be structured as follows:

• The top level are: 'groups', 'dimensions', 'compounds', 'variables', 'attrs_global' and 'attrs_groups'.

• 'attrs_global' and 'attrs_groups' are added in version 0.3.0

• The names of the attributes, groups, dimensions, compounds and variable should be specified as PosixPaths, however:

  • The names of groups should never start with a slash (always relative to root);
  • All other elements which are stored in root should also not start with a slash;
  • Hoewever the non-group elements require a starting slash (absolute paths) when they are stored not the root.

• The section 'groups' are optional, but you should provide each group you want to use in your file. The 'groups' section in the YAML file may look like this:

  groups:
    - engineering_data
    - image_attributes
    - navigation_data
    - science_data
    - processing_control/input_data
  

• The section 'dimensions' is obligatory, you should define the dimensions for each variable in your file. The 'dimensions' section may look like this:

  dimensions:
    days:
      _dtype: u4
      _size: 0
      long_name: days since 2024-01-01 00:00:00Z
    number_of_images:             # an unlimited dimension
      _dtype: u2
      _size: 0
    samples_per_image:            # a fixed dimension
      _dtype: u4
      _size: 307200
    /navigation_data/att_time:    # an unlimited dimension in a group with attributes
      _dtype: f8
      _size: 0
      _FillValue: -32767
      long_name: Attitude sample time (seconds of day)
      calendar: proleptic_gregorian
      units: seconds since %Y-%m-%d %H:%M:%S
      valid_min: 0
      valid_max: 92400
    n_viewport:                   # a fixed dimension with fixed values and attributes
      _dtype: i2
      _size: 5
      _values: [-50, -20, 0, 20, 50]
      long_name: along-track view angles at sensor
      units: degrees
  

• The 'compounds' are optional, but you should provide each compound data-type which you want to use in your file. For each compound element you have to provide a list with its data-type and optionally its number of elements. The 'compound' section may look like this:

  compounds:
    stats_dtype:
      time: [u8]
      index: [u2]
      tbl_id: [u1]
      saa: [u1]
      coad: [u1]
      texp: [f4]
      lat: [f4]
      lon: [f4]
      avg: [f4]
      unc: [f4, [2,]]
      dark_offs: [f4]
  

• The 'variables' are defined by their data-type ('_dtype') and dimensions ('_dims'), and optionally chunk sizes ('_chunks'), compression ('_compression'), variable length ('_vlen'). In addition, each variable can have as many attributes as you like, defined by its name and value. The 'variables' section may look like this:

  variables:
    /science_data/detector_images:
      _dtype: u2
      _dims: [number_of_images, samples_per_image]
      _compression: 3
      _FillValue: 65535
      long_name: Detector pixel values
      coverage_content_type: image
      units: '1'
      valid_min: 0
      valid_max: 65534
    /image_attributes/nr_coadditions:
      _dtype: u2
      _dims: [number_of_images]
      _FillValue: 0
      long_name: Number of coadditions
      units: '1'
      valid_min: 1
    /image_attributes/exposure_time:
      _dtype: f8
      _dims: [number_of_images]
      _FillValue: -32767
      long_name: Exposure time
      units: seconds
    stats_163:
      _dtype: stats_dtype
      _dims: [days]
      _vlen: True
      comment: detector map statistics (MPS=163)
  

Notes and ToDo

• The layout of a HDF5 or netCDF4 file can be complex. From version 0.3, you can split the file definition over several YAML files and provide a list with the names of YAML files as input to H5Yaml and NcYaml.
• From version 0.4, the classes H5Yaml and NcYaml are replaced by NcFromYaml. You can use module h5py to write the netCDF4 file or netCDF4 using the keyword 'module'. Then the classes H5Create or NcCreate perform the dict to netCDF4 conversion.

Support [TBW]

Road map

• Release v0.1 : stable API to read your YAML files and generate the HDF5/netCDF4 file

Authors and acknowledgment

The code is developed by R.M. van Hees (SRON)

License

• Copyright: Richard van Hees (SRON) (https://www.sron.nl).
• License: Apache-2.0