Holds information about values of a measurement of some scalar magnitude (namely, a
feature or a coordinate) over samples. Each variable has a range of possible values
which can be numerical or verbal descriptions. Variables are classified by their
, according to the relationships between their possible values. A
variable is nominal
if there are no relationships between the possible values of
the variable, which are typically non-numerical. For example, a wine primary taste
can be described as citrus, woody, or fruity. A variable is ordinal
values have an order. For example, a person can be described as thin, normal and fat.
Finally, a variable is numerical
if its values are numerical. We do not make a
discrimination between numerical values that are on a ratio scale (e.g., a person's height)
and those that are not (e.g., temperature in Celsius).
If the values are verbal descriptions, they are represented as integers, and the connection
between them and the verbal descriptions are kept in the lookup table lut
. The lut
field is therefore always defined for nominal variables, always not defined for numerical
variables, and sometimes defined for ordinal variables.
If the variable is known to be distributed according to some probability distribution,
this distribution is designated in the field distribution
. This field is a structure
with two fields - the name of the distribution (e.g., 'normal') and its parameters (e.g,
mean and variance). Statistical measures of the variable, namely mean
, are also kept. Each of these is a structure with a
'population' field and with 'sample' field, keeping the population and sampling measures,
respectively. Unknown population measures are designated by NaNs.
Someimes, a transformation is applied to a variable (e.g., log-transform, centering,
standardizing), yielding a related, but new, variable. In order to be able to track the
series of transfomations performed, these are recorded as the transformations
vector. Each element is a structure with a name field (e.g., 'center') and a parameter
field (e.g., one scalar for centering).
Navigate to: General Description