The Prosection Matrix: A tool to support the interactive exploration of statistical models and data (Q1297833)

The problem of multidimensional response optimization is considered by a set of parameters whose value the designer is free to choose. The dependence of the response \((S_1,\dots,S_m)\) on parameters \((X_1,\dots,X_d)\) is considered to be known. Two interactive visual tools are proposed for this purpose: the Influence Explorer and the Prosection Matrix. In the Influence Explorer for a set of random parameter vectors the responses are computed and then the panel of linked gistograms for \(X_1\), \dots, \(X_d\), \(S_1\), \dots, \(S_m\) values is displayed. The designer can select a subset of the full dataset using some restrictions on \(X_i\) and \(S_j\) interactively on the panel. The gistograms for subsets are displayed in different colors. Choosing the subset of parameters at which the responses lie in appropriate bounds solves the problem. The Prosection Matrix is a similar tool using two-dimensional density plots instead of gistograms. Some restrictions on the responses are fixed. For fixed values of \(X_i\) and \(X_j\) the values of \(X_k\), \(k\not=i\), \(k\not=j\), are taken at random and the values of responses \(S_1^*\), \dots, \(S_m^*\) are computed by the given model. The color of the point \((X_i,X_j)\) on the plot depends on the number of the restrictions satisfied by \(S_1^*\), \dots, \(S_m^*\). In the panel the matrix of all \((X_i,X_j)\), \(i,j=1,\dots,d\), plots is displayed. Defining different restrictions on the responses and parameters interactively the designer can choose the optimal values of parameters. A similar technique based on the parallel coordinates plots is also considered.