weight.entropy: Calculates decision weights using the entropy method

In avilesd/productConfig: Tools to analyze behavioral patterns from Product Configurators

Description

This function first normalizes a list of matrices and then calculates the decision weight for each attribute, using an entropy approach [1, 2], which can be categorised as an objective approach, just as weight.differenceToIdeal. This type of weight functions use only the information within the decision matrix to calculate weights. It does not need information about the decision maker's preferences.

Usage

weight.entropy(dataset, userid = NULL, attr = NULL, rounds = "all",
  cost_ids = NULL)

Arguments

dataset	data.frame with the user generated data from a product configurator. See decisionMatrix for specifications of the dataset.
userid	a vector of integers that gives the information of which users the matrix should be calculated. Vectorised.
attr	attributes IDs, vector of integer numbers corresponding to the attributes you desire to use; attr are assumed to be 1-indexed. This function will calculate with all attributes and do the subsetting a posteriori. If you want to get the weights for only two attributes you will have to first use decisionMatrix and then pass it on to normalize.sum and entropy.
rounds	integer vector or text option. Which steps of the configuration process should be taken into account? Defaults are "all" in order to have more data to calculate with. If "first" or "last" are entered there will be only one rounds to gather data from, consequently all attribtues will have the same weight.
cost_ids	argument used to convert selected cost attributes into benefit attributes. Integer vector. This functions uses an alternative normalizing method normalize.sum that does not produce negative values. However, it has one limitation, if within a same attribute there are negative and positive values, the function will likely produce a NaN.

Details

The result is a list of vectors, each vector with the same length as the number of columns of the input matrices, i.e. each column gets a weight. The sum of a weight vector should always equal 1.

In contrast to weight.differenceToIdeal, this function distributes lower weights to those attributes, which have similar values throughout the decision matrix.

Note: the normalizing function used here normalize.sum has one limitation, if within a same attribute there are negative and positive values, the function will likely produce a NaN.

cost_ids As in the other functions, if you enter a cost_ids that is not in your entered attributes, the functions will calculate the output with all attributes in your data, including the cost(s) attributes and only after the calculations does the function subset the result according to the attr. When the attributes and cost_ids differ, the function allows the calculation but it will throw a warning.

Value

a list of weight vector(s)

References

[1]Hwang, C. L., & Yoon, K. (2012). Multiple attribute decision making: methods and applications a state-of-the-art survey (Vol. 186). Springer Science & Business Media.

[2]Shannon, C. E. (2001). A mathematical theory of communication. ACM SIGMOBILE Mobile Computing and Communications Review, 5(1), 35.

[3]Lotfi, F. H., and Fallahnejad, R. (2010). Imprecise Shannons entropy and multi attribute decision making. Entropy, 12(1), 53-62.

Examples

# Not runnable yet
weight.entropy(myData, 15:22)
weight.entropy(laptop_data, 40:45, attr= c(1, 3, 4), cost_ids = 4)
savedWeights <- weight.entropy(myData, c(6, 15, 18, 20, 26), attr = 1:4, cost_ids = 4, rounds=1:8)
lapply(savedWeights, sum) # Should return 1 for any output of this function

avilesd/productConfig documentation built on May 11, 2019, 4:08 p.m.