Plugin Design - Regressions

Description

Draco includes a specialized class to aid in creating regression plugins. The class, Plugin::RegressionPluginSkeleton, is meant to simplify the generation of results for regressions which conform to the simple regression attributes:

Coefficients with standard errors and t-scores are calculated
A covariance matrix is computed
Coefficient names are based on variable names

For example, the results would be applicable for oridinary least squares or probit MLE calculations.

Necessary Attributes

Regression plugins are expected to assign values to attributes during calculations for results generation. The attributes are defined as follows:

Attribute	Expected Type	Description
@coef_names	array of strings	An array of strings that define the display names for each calculated coefficient (see below)
@covariance	Matrix	A matrix representing the covariance matrix resulting from the regression
@failure	string or nil	A string if the regression failed containing the reason, or nil otherwise
@statistics	array of arrays	An array of two-element arrays containing a statictic name and a value (see below)
@stderr	Vector	A vector representing the standard error of the coefficients
@tscores	Vector	A vector representing the t-scores for each coefficient
@x	Vector	A vector containing the estimates of the coefficients

Assigning Coefficient Names

Coefficient names can be automatically generated using a call to the generate_coef_names method. The method accepts an array of strings, presumably independent variable names, and generates appropriate coefficient names. The method also calls the method has_constant? to determine if a coefficient name of "Constant" should be included; the has_constant? method should be overridden appropriately to return a boolean.

Assigning Statistics

Statistics can be added to results via calls to the statistic method. The method is used as follows:



statistic name, value

The name should be the plain text name of the statistic to display, and value should be a numeric result.

Plugin Design - Regressions

Description

Necessary Attributes

Assigning Coefficient Names

Assigning Statistics

See Also