How to write a normalizer

Introduction

A normalizer can be any Python algorithm that takes the archive of an entry as input and manipulates (usually expands) the given archive. This way a normalizer can add additional sections and quantities based on the information already available in the archive.

All normalizer are executed after parsing. Normalizer are run for each entry (i.e. each set of files that represent a code run). Normalizer are run in a particular order and you can make assumptions about the availability of data created by other normalizer. A normalizer is run in any case, but it might chose not to do anything. A normalizer can perform any operation on the archive, but in general should not alter existing information, but just add more information.

Starting example

This is an example for a very simple normalizer that computes the unit cell volume from a given lattice and adds it to the archive.

from nomad.normalizing import Normalizer
from nomad.atomutils import get_volume

class UnitCellVolumeNormalizer(Normalizer):
    def normalize(self):
        for system in self.archive.section_run[-1].section_system:
            system.unit_cell_volume = get_volume(lattice_vectors.magnitude)

            self.logger.debug('computed unit cell volume', system_index=system.m_parent_index)

You simply inherit from Normalizer and implement the normalize method. The archive is available as a field. There is also a logger on the object that can be used. Be aware that the processing will already report the run of the normalizer, log its execution time, log any exceptions that might been thrown.

Of course, if you add new information to the archive, this needs also be defined in the metainfo. For example you could extend the section system with a special system definition that extends the existing section system definition:

import numpy as np
from nomad.datamodel.metainfo.public import section_system as System
from nomad.metainfo import Section, Quantity

class UnitCellVolumeSystem(System):
    m_def = Section(extends_base_section=True)
    unit_cell_volume = Quantity(np.dtype(np.float64), unit='m^3')

Or you simply alter the section_system class (nomad/datamodel/metainfo/public.py).

System normalizer

There is a special base-class for normalizing systems that allows to run the normalization on all (or only the resulting representative system:

from nomad.normalizing import SystemBasedNormalizer
from nomad.atomutils import get_volume

class UnitCellVolumeNormalizer(SystemBasedNormalizer):
    def _normalize_system(self, system, is_representative):
        system.unit_cell_volume = get_volume(lattice_vectors.magnitude)

The parameter is_representative will be true for the representative system, i.e. the final step in a geometry optimization or other workflow.

Adding a normalizer to the processing

For any new normalizer class to be recognized by the processing, the normalizer class needs to be added to the list of normalizers in nomad/normalizing/__init__.py. The order of the normalizers in this list will also determine the execution order of the normalizers during processing.

normalizers: Iterable[Type[Normalizer]] = [
    SystemNormalizer,
    UnitCellVolumeNormalizer,
    OptimadeNormalizer,
    DosNormalizer,
    BandStructureNormalizer,
    EncyclopediaNormalizer,
    WorkflowNormalizer
]

Testing a normalizer

To simply tryout a normalizer, you could use the CLI and run the parse command:

nomad --debug parse --show-archive <path-to-example-file>

But eventually you need to add a more formal test. Place your pytest-tests in tests/normalizing/test_unitcellvolume.py similar to the existing tests. Necessary test data can be added to tests/data/normalizers.