Greedy search

To identify this set of final feasible solutions, X e 1, with low scores, we developed a greedy search procedure, S (Saraiva and Stephanopoulos, 1992c), that has resulted, within an acceptable computation time, in almost-optimal solutions for all the cases studied so far, while avoiding the combinatorial explosion with the number of (x, y) pairs of an exhaustive enumeration/evaluation of all feasible alternatives. The algorithm starts by partitioning the decision space into a number of isovolu-... 

The search or optimization method, the third component, describes the computational procedure to search over parameters and structures. Issues here include the computational methods used to optimize the scoring function and to search related parameters such as the maximum number of iterations or convergence specifications for iterative algorithms. Typical search methods are greedy search, gradient-depen-dent search methods, or breadth search methods [4], One distinguishes between searches that involve only the optimization of the parameters in fixed structures, and the optimization of structures and parameters for data mining methods that include searches over parameter and structure. 

The full combinatorial search (Tables 7.9 and 7.10) confirms the superior model quality of M normalized indices, plus A index, already found with the greedy search (Tables 7.7 and 7.8), even if it does not need any empirical parameters to improve the model quality. The zero-level description shown in Tables 7.11 and 7.12 is not... 

Constructive searches are often ctdled greedy or myopic because they usutdly choose variables and values to fix on the basis of estimates of immediate or short-term gain. The main issue in design of such procedures is to choose an informative measure of efficiency or gain on which to base selections. 

Constructive searches based on more complicated efficiency ratios are much more common than the pure greedy notion of considering only objective coefficient magnitudes. One example is Dobson s (1982), heuristic for generalized covering problems of the form... 

In this process, T variables are randomly chosen from the set of possible variable choices and the variable with the highest pheromone value is selected. T is the size of the tournament. By varying T the greediness of the algorithm can be modified, lower values of T approximate random search as the competition element of the tournament is lessened and the influence of paths with high pheromone is reduced. Higher values of T increase the greediness of the search. 

Both search algorithms find a highly compact semiempirical combination shown in Tables 7.7 and 7.8 (greedy) and 7.9 and 7.10 (full combinatorial). Here, descriptor Tz/m = [2 and ° j/E belong to Kp(p-odd) / fs configuration]. The... 

The greedy descriptor is shown in Tables 7.7 and 7.8. The full combinatorial search algorithm (Tables 7.9 and 7.10) finds many similar semiempirical descriptors throughout the many types of configurations, among which the descriptor of Tables 7.9 and 7.10 has been chosen. The correlation vector to obtain Fig. 7.7 is,... 

The greedy (Tables 7.7 and 7.8) and complete combinatorial search algorithm (Tables 7.9 and 7.10) consider acetonitrile a strong outlier and take advantage of the two-valued (f>(P, 1) symmetry parameter that zeroes all indices whose properties have /t = 0, while leaves them unchanged if /t 0, i.e., for /t = 0, 4>-x = 0, while for II 0, (p-x = X- Th optimal full combinatorial descriptor with an improved r but is shown in Tables 7.9 and 7.10, while in Tables 7.11 and 7.12 is shown the corresponding zero-level description, which seems not excessively bad even if the aid of the non-random (f> ad hoc index plays a no minor role. The correlation vector of descriptor in Table 7.9, with whom Fig. 7.11 has been obtained, is,... 

The search encompassing a combinatorial space made of the rl- r38 and then (j)rdl-(l>rd38 plus 0MCI and Z and 4>Tx/m are cOTifiguration-dependent). 

The search for a good four-index semi-random descriptor done among MCI plus M, plus rl-r38 and then rdl-rd38 finds the optimal mc-exp-rn-AeEcnptov in Tables 7.13 and 7.14. This semi-random descriptor is statistically similar to the highly convoluted greedy descriptor but formally more down-to-earth. Without random index r30 the statistics are = 0.788, s = 0.05, q = 0.729 (here only is... 

All different MC or empirical indices of the full normal (non semi-random) descriptors of Tables 7.9 and 7.10 are now joined together to form a new space of indices, kind of super-indices, which will be used for a full combinatorial search of the best super-descriptors in a kind of configuration interaction of best indices. This super-descriptor space gave no remarkable results with the greedy algorithm, but it does find improved descriptions for the following four properties. 

---