Skip to content

How to write a YAML schema package

This guide explains how to write and upload NOMAD schema packages in the YAML format that can be uploaded as part of your data. This is a good way to start out experimenting with custom data structures in NOMAD, but for more advanced use cases you may need to use Python schema packages. For more information on how an archive file is composed, visit Explanation > Data structure.

Example data

Let's assume we want to describe chemical compositions using the elements they contain. The following structured data (in this example as a .yaml document) could describe the composition of water.

composition: H2O
elements:
- label: H
  density: 8.375e-05
  isotopes: [1, 2, 3]
- label: O
  density: 1.141
  isotopes: [16, 17, 18]

In structured data formats (such as .yaml or .json), data is put into combinations of primitive values (e.g. 'H2O', 1.141), objects (a set of keys and value pairs, where values can be objects, lists, or primitive values), and lists of values.

Sections

In a schema package, we want to describe the structure of data, i.e. what are the allowed combinations of objects, lists, and primitive values. The crucial task here is to define what keys certain types of objects can have and what possible values might exist for each of these keys.

In NOMAD, we call objects sections and we define types of objects with section definitions. Since objects can be nested, sections become like the sections and subsections of a book or paper. Sections are a representation of data and they are the building blocks for archives. Section definitions form a schema package and they are the building blocks for the metainfo.

In the above example, we have two types of objects: an overaching object for the entire structure (with keys for composition and elements), and an additional object which describes the internal structure of elements (with keys for label, density, and isotopes). Let's start with the definition for elements. This is what the section definition looks like in NOMAD's yaml-based schema package format:

Element:
  quantities:
    label:
      type: str
    density:
      type: np.float64
      unit: g/cm**3
    isotopes:
      type: int
      shape: ['*']

A section definition provides all the available keys for a section that instantiates this definition. For each key, e.g. label, density, isotopes, it provides more information on the possible values.

Let's have a look at the overall definition for our chemical composition:

Composition:
  quantities:
    composition:
      type: str
  sub_sections:
    elements:
      section: Element
      repeats: true

Again, all possible keys (composition and elements) are defined. But now we see that there are two different types of keys, quantities and subsections. We say that section definitions can have properties (e.g. the keys they define) and there are two distinct types of properties.

Quantities

Quantities define possible primitive values. The basic properties that go into a quantity definition are:

  • type: what kind of primitive value can be used, e.g. str or np.float64
  • shape: what is the shape of the value, e.g. scalar or list (['*'])
  • unit: what is the physical meaning of the value

The names of quantity definitions serve as the key, used in respective section objects.

Type

This is a list of supported quantity types.

type description
string
str
float
integer
int
boolean
bool
np.int32 Numpy based integer with 32 bits.
np.int64 Numpy based integer with 64 bits.
np.float32 Numpy based float with 32 bits.
np.float64 Numpy based float with 64 bits.
Datetime
User A type for NOMAD users as values.
Author A complex type for author information.
{type_kind: Enum, type_data: []} Use type_data to specify enum values as list of strings.
To define a quantity that is a reference to a specific section.

Shape

The shape of a quantity is a list of dimensions, where each dimension defines the possible size of that dimension. The empty list (or no shape) describes a scalar value, a list with one dimension a list or vector, a list with two dimensions a matrix, etc.

Dimensions can be given as:

  • an integer number to define a fixed size, e.g. a 3x3 matrix would have shape [3, 3].
  • the string '*' to denote am arbitrary sized dimension, e.g. a list quantity would have shape ['*'].
  • A string that describes the name of a sibling quantity with an integer type, e.g. ['number_of_atoms', 3]

Unit

NOMAD manages units and data with units via the Pint Python package. A unit is given as a string that is parsed by pint. These strings can be simple units (or their aliases) or complex expressions. Here are a few examples: m, meter, mm, millimeter, m/s, m/s**2.

While you can use all kinds of units in your uploaded schema packages, the built-in NOMAD schema (Metainfo) uses only SI units.

Subsections

Subsections define a part-of-relationship between two sections. Subsection definitions are properties of the parent section definition and name a child section definition. In the data, we can now contain instances of the target (e.g. Element) in instances of the source (e.g. Composition). A subsection can be defined as repeating to allow many child sections of the same type. In our example, one Composition can contain many Elements.

The names of subsection definitions serve as the key, used in respective section objects.

Uploading schema packages

NOMAD archive files allow you to upload data in NOMAD's native file format. An archive file can be a .yaml or .json file. It ends with .archive.json or .archive.yaml. Archive files are mainly used to convey data. Since YAML schema packages are also "just" data, archive files can also be used to convey a schema package.

You can upload schema packages and data in separate files. schema_package.archive.yaml

definitions:
  sections:
    Element:
      quantities:
        label:
          type: str
        density:
          type: np.float64
          unit: g/cm**3
        isotopes:
          type: int
          shape: ['*']
    Composition:
      quantities:
        composition:
          type: str
      sub_sections:
        elements:
          section: Element
          repeats: true

and data.archive.yaml

data:
  m_def: '../upload/raw/package.archive.yaml#Composition'
  composition: 'H2O'
  elements:
    - label: H
      density: 0.00008375
      isotopes: [1, 2, 3]
    - label: O
      density: 1.141
      isotopes: [16, 17, 18]

Or, you can upload the schema package and data in the same file:

definitions:
  sections:
    Element:
      quantities:
        label:
          type: str
        density:
          type: np.float64
          unit: g/cm**3
        isotopes:
          type: int
          shape: ['*']
    Composition:
      quantities:
        composition:
          type: str
      sub_sections:
        elements:
          section: Element
          repeats: true
data:
  m_def: Composition
  composition: H2O
  elements:
  - label: H
    density: 8.375e-05
    isotopes: [1, 2, 3]
  - label: O
    density: 1.141
    isotopes: [16, 17, 18]

References

Reference quantities

We already saw that we can define a part-of relationship between sections. When we want to represent highly inter-linked data, this is often insufficient. References allow us to create a more lose relationship between sections.

A reference is a uni-directional link between a source section and a target section. References can be defined in a schema package as a quantity in the source section definition that uses the target section definition as a type.

Instead of connecting the elements in a composition with subsections, we can also connect a composition section to elements with a quantity:

Composition:
  quantities:
    composition:
      type: str
    elements:
      type: Element
      shape: ['*']

Here, type: Element refers to the section definition Element, very similar to section: Element in a subsection definition.

We saw above that subsections are represented as nested objects in data (forcing a part-of relationship). References are represented as string-typed primitive values in serialized data. Here is an example Composition with references to elements:

composition: H2O
elements: ['#/data/periodic_table/elements/0', '#/data/periodic_table/elements/1']

These string-references determine the target section's place in the same archive. Each /-separated segment represents a key. A reference starts from the root object and following the sequence of keys to a specific object (i.e. section). Here is the full archive data:

data:
 periodic_table:
   elements:
   - label: H
     density: 8.375e-05
     isotopes: [1, 2, 3]
   - label: O
     density: 1.141
     isotopes: [16, 17, 18]
 compositions:
 - composition: H2O
   elements: ['#/data/periodic_table/elements/0', '#/data/periodic_table/elements/1']

If you follow the keys data, periodic_table, elements, 0, you reach the section that represent hydrogen. Keep in mind that lists use index-numbers as keys.

Schema package references

References can look different depending on the context. Above we saw simple references that point from one data section to another. But, you also already a saw a different type of reference. Schema packages themselves contain references: when we used type: Element or section: Element to refer to a section definition, we were writing down references that point to a section definition. Here we can use a convenience representation: Element simply replaces the otherwise cryptic #/definitions/sections/0.

So far, we never discussed the use of m_def. In the examples you might have seen this as a special key in some objects. Whenever we cannot determine the section definition for a section by its context (e.g. the key/subsection used to contain it in a parent section), we use m_def to provide a reference to the section definition.

Different forms of references

Depending on where references are used, they might take a different serialized form. Here are a few examples for different reference syntax:

Example reference Comments
#/data/periodic_table/elements/0 Reference to a section within the subsection hierarchy of the same archive.
Element Reference to a section definition in the same archive. Can only be used to target section definitions.
nomad.datamodel.metainfo.workflow Reference to a section definition that was written in Python and is part of the NOMAD code. Can only be used to target section definitions.
../upload/raw/data.archive.yaml#/data Reference to a section in a different .archive.yaml file of the same upload.
../upload/archive/mainfile/data.archive.yaml#/data Reference to a section in a processed archive given by entry mainfile.
../upload/archive/zxhS43h2kqHsVDqMboiP9cULrS_v#/data Reference to a section in a processed archive given by entry-id.
../uploads/zxhS43h2kqHsVDqMboiP9cULrS_v/raw/data.archive.yaml#/data Reference to a section in an entry of a different upload.
https://mylab.eu/oasis/api/v1/uploads/zxhS43h2kqHsVDqMboiP9cULrS_v/raw/data.archive.yaml#/data Reference to a section in an entry in a different NOMAD installation.

References across entries

A references in the archive of one entry can point to a section in a different entry's archive. The following two example files, exemplify this use of reference between two NOMAD entries.

periodic_table.archive.yaml

definitions:
  sections:
    Element:
      quantities:
        label:
          type: str
        density:
          type: np.float64
          unit: g/cm**3
        isotopes:
          type: int
          shape: ['*']
    PeriodicTable:
      sub_sections:
        elements:
          repeats: true
          section: Element
data:
  m_def: PeriodicTable
  elements:
  - label: H
    density: 0.00008375
    isotopes: [1, 2, 3]
  - label: O
    density: 1.141
    isotopes: [16, 17, 18]

composition.archive.yaml

definitions:
  sections:
    Composition:
      quantities:
        composition:
          type: str
        elements:
          type: ../upload/raw/periodic_table.archive.yaml#Element
          shape: ['*']
data:
  m_def: Composition
  composition: 'H2O'
  elements:
    - ../upload/raw/periodic_table.archive.yaml#data/elements/0
    - ../upload/raw/periodic_table.archive.yaml#data/elements/1

These inter-entry references have two parts: <entry>#<section>, where entry is a path or URL denoting the target entry and section a path within the target entry's subsection containment hierarchy.

Please note that also schema packages can be spread over multiple files. In the above example, one file contained the schema package and data for a periodic table and another file contained schema package and data for the composition of water (using the periodic table).

Base sections and inheritance

We add a relationship between section definitions that allows us to create more specialized definitions from more abstract definitions. Here the properties of the abstract definition are inherited by the more specialized definitions

Base sections

Here is a simple schema package with two specialization of the same abstract section definition:

definitions:
  sections:
    Process:
      quantities:
        time:
          type: Datetime
    Evaporation:
      base_section: Process
      quantities:
        pressure:
          type: np.float64
          unit: Pa
    Annealing:
      base_section: Process
      quantities:
        temperature:
          type: np.float64
          unit: K

The two specialized definitions Annealing and Evaporation define the abstract definition Process via the base_section property. With this Annealing and Evaporation inherit the quantity time. We do not need to repeat quantities from the base section, and we can add more properties. Here is an example Evaporation using both the inherited and added quantity:

data:
  m_def: Evaporation
  time: '2022-10-13 12:00:00'
  pressure: 100

Polymorphy

What happens if we reference abstract definitions in subsections or reference quantities? Here is an subsection example. In one schema, we define the relationship between Sample and Process. In another schema, we want to add more specializations to what a process is.

abstract.archive.yaml

definitions:
  sections:
    Process:
      quantities:
        time:
          type: Datetime
    Sample:
      sub_sections:
        processes:
          section: Process
          repeats: true

specialized.archive.yaml

definitions:
  sections:
    Evaporation:
      base_section: ../upload/raw/abstract.archive.yaml#Process
      quantities:
        pressure:
          type: np.float64
          unit: Pa
    Annealing:
      base_section: ../upload/raw/abstract.archive.yaml#Process
      quantities:
        temperature:
          type: np.float64
          unit: K

The section definition use in the subsection processes defines what a contained section has to be "at least". Meaning that any section based on a specialization of Process would be a valid processes subsection.

specialized.archive.yaml

definitions:
  # see above
data:
  m_def: ../upload/raw/abstract.archive.yaml#Sample
  processes:
  - m_def: Evaporation
    time: '2022-10-13'
    pressure: 100
  - m_def: Annealing
    time: '2022-10-13'
    temperature: 342

The fact that a subsection or reference target can have different "forms" (i.e. based on different specializations) is called polymorphism in object-oriented data modelling.

Pre-defined sections

NOMAD provides a series of built-in section definitions. For example, there is EntryArchive, a definition for the top-level object in all NOMAD archives (e.g. .archive.yaml files). Here is a simplified except of the main NOMAD schema nomad.datamodel:

EntryArchive:
  sub_sections:
    metadata:
      section: EntryMetadata
    definitions:
      section: nomad.metainfo.Package
    data:
      section: EntryData
    # ... many more
EntryData:
  # empty

Compare this to the previous examples: we used the top-level keys definitions and data without really explaining why. Here you can see why. The EntryArchive property definitions allows us to put a schema package into our archives. And the EntryArchive property data allows us to put data into archives that is a specialization of Schema. The Schema definition is empty. It is merely an abstract placeholder that allows you to add specialized data sections to your archive. Therefore, all section definitions that define a top-level data section, should correctly use nomad.datamodel.Schema as a base section. This would be the first "correct" example:

definitions:
  sections:
    Greetings:
      base_section: nomad.datamodel.EntryData
      quantities:
        message:
          type: str
data:
  m_def: MyData
  message: Hello World

Here are a few other built-in section definitions and packages of definitions:

Section definition or package Purpose
nomad.datamodel.EntryArchive Used for the root object of all NOMAD entries
nomad.datamodel.EntryMetadata Used to add standard NOMAD metadata such as ids, upload, processing, or author information to entries.
nomad.datamodel.EntryData An abstract section definition for the data section.
nomad.datamodel.ArchiveSection Allows to put normalize functions into your section definitions.
nomad.datamodel.metainfo.eln.* A package of section definitions to inherit commonly used quantities for ELNs. These quantities are indexed and allow specialization to utilize the NOMAD search.
nomad.parsing.tabular.TableData Allows to inherit parsing of references .csv and .xls files.
nomad.datamodel.metainfo.workflow.* A package of section definitions use by NOMAD to define workflows
nomad.metainfo.* A package that contains all definitions of definitions, e.g. NOMAD's "schema language". Here you find definitions for what a sections, quantity, subsections, etc. is.

Separating data and schema package

As we saw above, a NOMAD entry can contain schema package definitions and data at the same time. To organize your schema package and data efficiently, it is often necessary to re-use schema packages and certain data in other entries. You can use references to spread your schema packages and data over multiple entries and connect the pieces via references.

Here is a simple schema package, stored in a NOMAD entry with mainfile name package.archive.yaml:

definitions:
  sections:
    Composition:
      quantities:
        composition:
          type: str
        base_composition:
          type: Composition
      sub_sections:
        elements:
          section: Element
          repeats: True
    Element:
      quantities:
        label:
          type: str
    Solution:
      quantities:
        solvent:
          type: Composition
      sub_sections:
        solute:
          section: Composition

Now, we can re-use this schema package in many entries via references. Here, we extend a schema contained in the package and instantiate definitions is a separate mainfile data-and-package.archive.yaml:

definitions:
  sections:
    SpecialElement:
      # Extending the definition from another entry
      base_section: '../upload/raw/package.archive.yaml#Element'
      quantities:
        atomic_weight:
          type: float
          unit: 'g/mol'
data:
  # Instantiating the definition from another entry
  m_def: '../upload/raw/package.archive.yaml#Composition'
  composition: 'H2O'
  elements:
    # Implicitly instantiate Element as defined for Composition.elements
    - label: H
    # Explicitly instantiate SpecialElement as a polymorph substitute
    - m_def: SpecialElement
      label: O
      atomic_weight: 15.9994

Here is a last example that re-uses the schema and references data from the two entries above:

definitions:
  sections:
    Composition:
      quantities:
        composition:
          type: str
        base_composition:
          type: Composition
      sub_sections:
        elements:
          section: Element
          repeats: True
    Element:
      quantities:
        label:
          type: str
    Solution:
      quantities:
        solvent:
          type: Composition
      sub_sections:
        solute:
          section: Composition

Attention

You cannot create definitions that lead to circular loading of *.archive.yaml files. Each definitions section in an NOMAD entry represents a schema package. Each schema package needs to be fully loaded and analyzed before it can be used by other schema packages in other entries. Therefore, two schema packages in two entries cannot reference each other.

Conventions

Conventions for labels

When assigning labels within your codebase, it's essential to follow consistent naming conventions for clarity and maintainability. The following guidelines outline the conventions for labeling different elements:

  • Sections: Labels for sections should adhere to Python convention of CapitalizedCamelCase. This means that each word in the label should begin with a capital letter, and there should be no spaces between words. For example: SectionLabelOne, SectionLabelTwo.

  • Quantities and Subsections: Labels for quantities and subsections should be in lower_case. This convention involves writing all lowercase letters and separating words with whitespace. Abbreviations within these labels may be capitalized to enhance scientific readability. For example: quantity label, subsection label, IV label.