Attribute types

Attribute types
Categorical:
Ordered:

Ordinal:
Quantitative:

Sequential:
Diverging:
Cyclic:

Attribute types: categorical

Categorical data (nominal) dont have any implicit ordering, but they
often have a hierarchical structure.
Categories distinguish if two items are the same (apples) or different
(pears vs. apples)
However, we can establish an external ordering upon categorical data.
For example, fruits can be alphabetically ordered, by the market
price, but only if additional information is available.

Attribute types: ordered

Ordered data have an implicit ordering.
Ordinal data can establish a well-defined ordering, but we can not
make mathematical operations with them. For example: size (S, M, L,
XL), rankings,
Quantitative data are measurements of magnitudes that suppor
mathematical operations. For example: height, weight, temperature,

In most cases, ordered data are quantitative.

Ordered attribute types

Ordered data are sequential if there is a range varying from a minimum to
a maximum value.
On the other side, ordered data are diverging if they can be divided in two
sequences pointing to opposite directions that share a common origin.
For example, altimetry and bathymetry are sequential magnitudes, but the
combination of both is a diverging measure.
Ordered data are cyclic if values return to a starting point instead of
increasing indefinitely.
For example, time magnitudes like the hour of the day, the day of the
week, the month of the year.

Hierarchical attributes
Combination of different basic attributes can be structured in a
hierarchical way.
There are many examples in the real-world. For example: time series
with the daily prices of stock. Different levels in the hierarchy
correspond to the resolution of the time series: days, weeks, months,
years,

Attribute semantics
Attribute semantics is the true meaning in the real-world.
The attribute type is its structural or mathematical interpretation.
The type of the attribute is not meaningful for its semantics, because
they are crosscutting questions.
Although there are several approaches for classifying attribute
semantics, in our case we will focus on discriminating keys vs. values;
spatial and continuous data vs. non spatial and discrete data;
identifying temporal attributes.

Key vs. value semantics

Key attributes are used as indices for looking up value attributes.
Key synonyms: independent attribute, dimension.
Value synonyms: dependent attribute, measure.
Distinction between key and value attributes is very significant for the
dataset types of tables, fields, and networks.

Flat table semantics

A flat table has only one key. Each item is a row in the table with any
number attributes as columns.
Keys can be implicit, meaning that its just the index of the row, or
explicit being contained within the table as an attribute.
In this case, there must not be any duplicate values within this
attribute.
Keys can be categorical or ordinal, but quantitative attributes are
typically unsuitable as keys because there is nothing to prevent them
from having the same values for multiple items.

Flat table semantics

Point Position

Point Position X

Point Position Y

Point Position Z

Unit

Name

Time

Object

Group

Index

150.

100.

122. Um

1 Measurement Points 4

Surpass Scene

150.

100.

122. um

1 Measurement Points 4

Surpass Scene

150.

99.

123. um

1 Measurement Points 4

Surpass Scene

150.

99.

122. um

1 Measurement Points 4

Surpass Scene

152.

99.

123. um

1 Measurement Points 4

Surpass Scene

152.

99.

123. um

1 Measurement Points 4

Surpass Scene

153.

97.

121. um

1 Measurement Points 4

Surpass Scene

153.

98.

121. um

1 Measurement Points 4

Surpass Scene

153.

97.

120. um

1 Measurement Points 4

Surpass Scene

153.

97.

121. um

1 Measurement Points 4

Surpass Scene

10

154.

96.

121. um

1 Measurement Points 4

Surpass Scene

11

154.

96.

121. um

1 Measurement Points 4

Surpass Scene

12

154.

100.

119. um

2 Measurement Points 4

Surpass Scene

13

154.

100.

120. um

2 Measurement Points 4

Surpass Scene

14

154.

101.

122. um

2 Measurement Points 4

Surpass Scene

15

154.

101.

122. um

3 Measurement Points 4

Surpass Scene

16

155.

100.

120. um

3 Measurement Points 4

Surpass Scene

17

The attribute Name is a

categorical attribute
candidate for being a key,
but there are several items
with the same value (O).
Quantitative attributes like
Point position, can also
appear duplicated in
several items.
In this example, we use the
last attribute, Index, as
the key table. This attribute
can be considered ordered
if the input order of the
items has a meaningful
semantic for representing
this temporal information,
or alternatively, it can be
considered categorical if it
just represents a unique
code.
Quantitative, ordered,
categorical.

Multidimensional table semantics

A more complex situation appears with a multidimensional table, where
multiple keys are used for accessing to the contents of one item.
The combination of all the keys must be unique for each item, although an
individual key attribute may contain duplicates.
The information about which attributes are keys and which ones arent
may not be available, and many times, determining which attributes are
independent (keys) vs. dependent is the goal of the visualization process
instead of the beginning of the analysis.
In this case, the result of the visual analysis would be to transform a flat
table into a more semantically meaningful multidimensional table.

Field semantics
In spatial fields, the spatial position is used as a quantitative key.
The main difference between fields and tables is that in the case of
fields, useful queries for attribute values are posed for locations
throughout the sampled range, not just the exact points where data
were recorded.
Fields can be characterized in terms of the number of keys versus the
number of values: the multivariate structure depends on the number
of value attributes, while the multidimensional structure depends on
the number of keys.

Field semantics: multidimensional structures

Most frequent multidimensional fields are 2D and 3D, where data is
collected from the static measures sampled over 2D and 3D
dimensions.
In the case that one of time-varying measures, these fields are
extended to 3D and 4D spaces.
Multidimensional structures can also be multivariate if it has multiple
keys and multiple values.
Multivariate structures can be classified as scalar, vector and tensor
fields.

Field semantics: scalar fields

Scalar fields are univariate, with only one value associated to each
point of the multidimensional space.
Geometrically, the intuition is that each point of the multidimensional
space has only one attribute.
A point in space can have several values associated with it. If there is
no relation between them, they are just several independent scalar
fields.
Examples: 3D medical scanning, temperature in a room at each point
in the 3D space.

Field semantics: vector fields

Vector fields are multivariate, with a collection of multiple values
associated to each location of the multidimensional space.
Geometrically, the intuition is that each point of the vector field has a
direction and a magnitude:
The dimensionality of the field defines the number of vector
elements (usually, 2, 3 or 4).
The vector length can be computed from the vector elements using,
for example, the Euclidean distance.
For example: the speed of the air in a room at some specific location
and time. For each item we measure its direction and speed
magnitude.

Field semantics: tensor fields

Tensor fields are multivariate fields with a more complex structure at
each location and with more attributes than vector fields.
Geometrically, the intuition is that each point of the vector field can
not be represented by one arrow and requires a more complex shape.
For example:
Example: the stress can be determined by nine values that represent
the acting forces in the three orthogonal directions.

Temporal semantics
A temporal attribute is just any kind of information related with the
concept time.
Temporal data are difficult to handle because the complex hierarchical
structure that we use to think about time and for the potential periodicity
of these data.
The nature of temporal hierarchy is multiscale: from nanoseconds to hours,
decades or centuries.
Analysis tasks involving temporal data require most often to discover or to
verify the data periodicity either at a predetermined scale or at some scale
not known in advance.
Furthermore, the temporal scales of interest maybe dont fit well into a
structured hierarchy: for example, weeks (7 days) and months (28, 30, 31
days).

Temporal attribute semantics

Temporal values can be considered as keys or as values.
Examples of temporal values: length of a period of time or date of a
transaction.
Examples of temporal keys: a medical scanner can have as
independent keys the dimensions x, y, z, t, and as value, the density
of the scanned tissue at a specific location and time.
Temporal attributes can be considered as quantitative values,
although they can also be considered ordered values if the time
passed between different events lacks of interest.

Time-varying datasets
A dataset is considered as time-varying if one of its key attributes is
the time.
Example: an animal tracking system with a fixed sampling frequency.
A very common case of temporal data are time-series datasets,
namely an ordered sequence of time-value pairs.
These datasets are a particular case of tables, where time is the key.
Usually, sampling intervals are uniform, but not always.
Task analysis of these datasets can be focused to find trends,
correlations or variations at different time scales.
In some contexts, the term dynamic is a synonym of time-varying
semantics. In our case, we will only consider in this way if the dataset
is a stream.

References
Tamara Munzner. Visualization Analysis and Design. A K Peters
Visualization Series. CRC Press. Oct 2014.
Stuart K. Card, Jock Mackinlay and Ben Shneiderman. Readings in
Information Visualization: Using Vision to Think. Morgan Kaufmann,
1999.