The HPDECIDE Procedure

names the input data set that contains the decision matrix or the prior probabilities, or both. This argument is required.

The named data set must contain the target variable that is specified in the TARGET statement. It might also contain decision variables that are specified in the DECVARS= option and prior probability variables that are specified in the PRIORVAR= option or the OLDPRIORVAR= option or both.

For a categorical target variable, there should be one observation for each class. Each entry in the decision matrix indicates the consequence of selecting target value i for variable j. If any class appears more than once in this data set, an error message is printed and the PROC HPDECIDE terminates. Any class value in the input data set that is not found in this data set is treated as a missing class value. The classes in this data set must correspond exactly to the variables in the POSTERIORS statement.

For an interval target variable, each row defines a knot in a piecewise linear spline function. The consequence of making a decision is computed by interpolation in the corresponding column of the decision matrix. If the predicted target value is outside the range of knots in the decision matrix, the consequence is computed by linear extrapolation. If the target values are monotonically increasing or decreasing, any interior target value is allowed to appear twice in data set. This enables you to specify discontinuities in the data. No target value is allowed to appear more than twice. If the target values are not monotonic, then they are sorted by PROC HPDECIDE and are not allowed to appear more than once.

The data set option type is specified in parentheses after the data set name when the data set is created or used. The possible values of type are LOSS, PROFIT, and REVENUE; the default is PROFIT.

specifies the numeric decision variables in the DECDATA= data set that contain the target-specific consequences for each decision. The decision variables cannot contain any missing values.

specifies the numeric constants that give the cost of a decision, the numeric variables in the input data set that contain case-specific costs, or any combination of constants and variables.

The number of cost constants and variables must match the number of decision variables in the DECVARS= option. You cannot use abbreviated variable lists. For any observation in which a cost variable is missing, the results for that observation are considered missing. By default, all costs are assumed to be 0. You can specify this option only when type is REVENUE.

specifies the number variable in the DECDATA= data set that contains the prior probabilities that are used to make decisions. Prior probabilities are also used to adjust the total and average profit or loss. Prior probabilities cannot be missing or negative, and there must be at least one positive prior probability. The prior probabilities are not required to sum to 1. But if they do not sum to 1, then they are scaled by some constant so that they do sum to 1. If you do not specify this option, then no adjustment for prior probabilities is applied to the posterior probabilities.

specifies the numeric variable in the DECDATA= data set that contains the prior probabilities that were used the first time the model was fit. If you specify this option, then you must also specify the PRIORVAR= option.