|
|
|
|
|
|
|
|
|
Orbfeq
| Глоссарий по искусственному интеллекту |
An ordinary radial basis function with equal widths on each node. see also: radial basis function, width, ftp://ftp.sas.com/pub/neural/faq2.html.
|
|
|
Ordinal attribute, английский
An ordinal attribute is one that takes on values whose order has external meaning but whose particular values or differences between values do not have meaning. an example is an unanchored rating by a person on a ninepoint scale. assigning one object a value of three and another a value of six does not imply that the second object has twice the value of the first. it only implies that it has more. by extension, you could also infer that the latter also has more than any other objects that are rated as 1, 2, 4, or 5 by the same person, and less than objects that are rated as 7, 8, or 9. the results of operations other than proportion level operations and order-specific counting (e.g., the number of cases with a score less than three) are dependent on the scaling, which is arbitrary. using techniques designed for continuous measures on ordinal variables often leads to misleading or silly results. see also: interval attribute, machine learning, logistic regression, nominal attribute (type), ratio attribute. ordinary least squares (ols) an ordinary least squares (ols) function uses the sum of the squared deviations between the observed and fitted values as its minimization criteria. this is equivalent to minimizing the euclidian distance between the pbserved and fitted values. variants include weighted least squares, which weights the individual squared differences according to a set of weights, least absolute deviations (lad) which minimize the absolute differences, and general lp criteria, where the p-th power of the absolute difference is minimized. ols and lad are l2 and l1, respectively. ols is known in the neural network literature as "least mean squares." the same acronym has been used in the neural network literature as orthogonal least squares a technique for forward stepwise selection in radial basis function (rbf) networks. the latter technique starts with a large set of candidate points and selects a subset that is useful for predictions. see also: radial basis function.
Orbfun, английский
An ordinary radial basis function with unequal widths on each node. see also: radial basis function, width, ftp://ftp.sas.com/pub/neural/faq2.html.
|
|
|
|
|
|
|
