Data Structures

The data structure for spectroscopic data is Orange.data.Table. Conversion to and from other data models is provided by ewoksxas.converters.orange.Converter.

Terminology

Domain	Row	Feature Column	Target Column	Meta Column	Description
Tabular data	row	column	column	column	A row is one record; a column is one field.
Machine learning	sample	feature	target	metadata	A sample is described by features and may have targets and metadata.
Statistics	observation	variable	response	covariate	An observation contains measured variables and associated information.
Measurement	measurement	variable	result	metadata	A measurement contains measured quantities and contextual information.
Orange	instance	feature	target	meta	An instance is one table row; columns have feature, target, or meta roles.
Spectroscopy	spectrum	channel	property	metadata	A spectrum contains intensities sampled along a spectral coordinate (e.g. photon energy).

Orange Table Structure

Each column has one of three roles:

• feature: measured values used as input data.

• target: values to predict or analyze.

• meta: additional information not used as input or target.

Example:

Filename	Scan number	Sample	7110 eV	7111 eV	Quality
meta	meta	meta	feature	feature	target
fe_foil.h5	1.1	Fe foil	0.842	0.861	good
fe_foil.h5	2.1	Fe foil	0.735	0.754	good
sample.h5	3.1	Catalyst A	0.612	0.645	poor

Features

Features contain the measured spectral signal. In XAS, feature names represent photon energy coordinates (typically in eV), and feature values represent the measured response at each energy.

Example:

• Feature names: [7100, 7101, 7102] eV

• Feature values: measured absorption signal at these energies

Targets

Targets are variables to predict or analyze. They are optional and mainly used in machine learning workflows.

Examples:

• Quality: sample quality score

• Class: material category (glass, metal, plastic)

• Concentration: estimated chemical concentration

Targets are represented by:

• Orange.data.ContinuousVariable for numeric values (e.g. concentration = 2.5 mg/L)

• Orange.data.DiscreteVariable for categories (e.g. class = leaf or root)

Metadata (metas)

Metadata describes each spectrum but is not used as input features or prediction targets.

Examples:

• sample location

• motor position

• acquisition timestamp

• experiment identifier

Metadata can be numeric, categorical, or text.