Difference between revisions of "Generalized Iterative Scaling"
| Line 1: | Line 1: | ||
| − | |||
| − | == | + | == The method == |
| − | The Generalized Iterative | + | The Generalized Iterative Scaling (GIS) is a [[Category::method]] that searches the exponential family of a Maximum Entropy solution of the form: |
| + | |||
| + | <math> | ||
| + | P(x) = \prod_i \mu_i ^{f_i(x)} | ||
| + | </math> | ||
| + | |||
| + | where the <math>\mu_i</math>'s are some unknown constants to be found. The <math>\mu_i</math>'s of the solution would be such that will make <math>P(x)</math> satisfy all the constraints <math>K_i</math>, of the equation: | ||
| + | |||
| + | <math> | ||
| + | \sum_x P(x)f_i(x) = K_i | ||
| + | </math> | ||
| + | |||
| + | == The Algorithm == | ||
| + | GIS starts with arbitrary <math>\mu_i ^{(0)}</math> values, wich define the initial probability estimate: | ||
| + | |||
| + | <math> | ||
| + | P^{(0)}(x) = \prod_i \mu_i ^{(0) f_i (x)} | ||
| + | </math> | ||
| + | |||
the circumstances under which it is meant to be used | the circumstances under which it is meant to be used | ||
| Line 10: | Line 27: | ||
and list papers that use it | and list papers that use it | ||
| − | |||
| − | |||
| − | |||
things the method is comparable to. | things the method is comparable to. | ||
| Line 19: | Line 33: | ||
| − | |||
| + | == Intrinsic characteristics == | ||
| + | GIS has three advantages when compared to other methods: it is able to incorporate feature selection, scales up well in numbers of features and is resilient to feature dependence. | ||
| − | + | On the other hand GIS has problems with smoothing and is relatively slow in training when compared to other classification methods | |
| − | |||
| − | |||
| − | |||
| − | |||
| − | |||
| − | |||
| − | methods | ||
| − | |||
| − | |||
| − | |||
== Related Papers == | == Related Papers == | ||
Revision as of 20:23, 29 September 2011
The method
The Generalized Iterative Scaling (GIS) is a method that searches the exponential family of a Maximum Entropy solution of the form:
where the 's are some unknown constants to be found. The 's of the solution would be such that will make satisfy all the constraints , of the equation:
The Algorithm
GIS starts with arbitrary values, wich define the initial probability estimate:
the circumstances under which it is meant to be used
you are expected to explain clearly what the method is
and list papers that use it
things the method is comparable to.
Explain what motivations or assumptions underlie the method
Intrinsic characteristics
GIS has three advantages when compared to other methods: it is able to incorporate feature selection, scales up well in numbers of features and is resilient to feature dependence.
On the other hand GIS has problems with smoothing and is relatively slow in training when compared to other classification methods
