Uters working in parallel. N 4 6 8 10 12 13 14 15 16 20 40 60 80 100 Total Solutions 15 4.7e+02 5.2e+04 1.5e+07 8.5e+09 2.6e+11 8.9e+12 3.5e+14 1.6e+16 1.7e+23 9e+65 5.1e+116 5.1e+172 4.4e+232 Seconds 0.00012 0.0037 0.4 1.1e+02 6.6e+04 2e+06 7e+07 2.8e+09 1.2e+11 1.3e+18 7e+60 4e+111 4e+167 3.4e+227 Years 3.7e-12 1.2e-10 1.3e-08 3.6e-06 0.0021 0.064 2.2 87 3.9e+03 4.2e+10 2.2e+53 1.3e+104 1.3e+160 1.1e+doi:10.1371/journal.pone.0124942.tcombination of sub-solutions which break the available assemblages into sets. When this possibility is included, the growth of possible solutions is even greater than factorial (Table 1). The combinatorial challenge with DFS has generally led many to use approximate approaches, based upon reduced similarity descriptions of type frequencies. Deterministic algorithms for frequency seriation, TSA side effects however, have advantages over similarity approaches since they make use of all of the type abundance information for each assemblage to build orders, thus allowing orders to be rejected and the search space thus reduced. Currently, hand-built approaches have been the only feasible way of creating deterministic seriation solutions [32, 33, 80]. In addition to integrating pairwise statistical evaluation for comparison of assemblages [32], manual solutions have the advantage of a general pattern recognition strategy that is inherent in our cognition. The disadvantage of hand-built solutions, even augmented by pairwise significance tests and bootstrap confidence intervals [32], is that investigators tend to stop when they find a valid solution given the effort involved. But a solution may be one of many possible, each representing potential information about change in cultural traits and their spatiotemporal histories. If what we seek is not merely a rough chronological order but information about cultural transmission, then we need to study all of the solutions. Ultimately, neither manual sorting nor probabilistic methods are satisfactory since the strength of seriation as a method rests on statistical assessment of all solutions that match the dual requirements of continuity and unimodality. Thus, an exhaustive characterization of the search space to find all of the valid orders is integral to the method. In addition, we need to know how sets of assemblages fail to produce a valid seriation order. Since we explain variability in frequencies as a function of transmission through time and space, finding the points at which assemblages cannot be fitted together is as important as finding those assemblages that can be seriated [32, 80]. In contrast, probabilistic orderings force all data points into a single solution, and thus are limited in their ability to locate the boundaries at which seriation solutions cannot be constructed. As a consequence, probabilistic seriation methods are generally unsuitable for disentangling the contributions of space, intensity of contact, and time.PLOS ONE | DOI:10.1371/journal.pone.0124942 April 29,7 /The IDSS Frequency Seriation AlgorithmFig 2, for example, demonstrates the kind of results that occur using correspondence analysis, which is the best available probability-based seriation technique [39, 69, 81, 82]. The Actinomycin D custom synthesis example is a set of assemblages of well-described ceramics from the lower Mississippi River Valley [10, 32, 83]. As shown in Panel A of Fig 2, the results generally meet the expectation of unimodality, but there are many deviations in the distribution. When we examine the distri.Uters working in parallel. N 4 6 8 10 12 13 14 15 16 20 40 60 80 100 Total Solutions 15 4.7e+02 5.2e+04 1.5e+07 8.5e+09 2.6e+11 8.9e+12 3.5e+14 1.6e+16 1.7e+23 9e+65 5.1e+116 5.1e+172 4.4e+232 Seconds 0.00012 0.0037 0.4 1.1e+02 6.6e+04 2e+06 7e+07 2.8e+09 1.2e+11 1.3e+18 7e+60 4e+111 4e+167 3.4e+227 Years 3.7e-12 1.2e-10 1.3e-08 3.6e-06 0.0021 0.064 2.2 87 3.9e+03 4.2e+10 2.2e+53 1.3e+104 1.3e+160 1.1e+doi:10.1371/journal.pone.0124942.tcombination of sub-solutions which break the available assemblages into sets. When this possibility is included, the growth of possible solutions is even greater than factorial (Table 1). The combinatorial challenge with DFS has generally led many to use approximate approaches, based upon reduced similarity descriptions of type frequencies. Deterministic algorithms for frequency seriation, however, have advantages over similarity approaches since they make use of all of the type abundance information for each assemblage to build orders, thus allowing orders to be rejected and the search space thus reduced. Currently, hand-built approaches have been the only feasible way of creating deterministic seriation solutions [32, 33, 80]. In addition to integrating pairwise statistical evaluation for comparison of assemblages [32], manual solutions have the advantage of a general pattern recognition strategy that is inherent in our cognition. The disadvantage of hand-built solutions, even augmented by pairwise significance tests and bootstrap confidence intervals [32], is that investigators tend to stop when they find a valid solution given the effort involved. But a solution may be one of many possible, each representing potential information about change in cultural traits and their spatiotemporal histories. If what we seek is not merely a rough chronological order but information about cultural transmission, then we need to study all of the solutions. Ultimately, neither manual sorting nor probabilistic methods are satisfactory since the strength of seriation as a method rests on statistical assessment of all solutions that match the dual requirements of continuity and unimodality. Thus, an exhaustive characterization of the search space to find all of the valid orders is integral to the method. In addition, we need to know how sets of assemblages fail to produce a valid seriation order. Since we explain variability in frequencies as a function of transmission through time and space, finding the points at which assemblages cannot be fitted together is as important as finding those assemblages that can be seriated [32, 80]. In contrast, probabilistic orderings force all data points into a single solution, and thus are limited in their ability to locate the boundaries at which seriation solutions cannot be constructed. As a consequence, probabilistic seriation methods are generally unsuitable for disentangling the contributions of space, intensity of contact, and time.PLOS ONE | DOI:10.1371/journal.pone.0124942 April 29,7 /The IDSS Frequency Seriation AlgorithmFig 2, for example, demonstrates the kind of results that occur using correspondence analysis, which is the best available probability-based seriation technique [39, 69, 81, 82]. The example is a set of assemblages of well-described ceramics from the lower Mississippi River Valley [10, 32, 83]. As shown in Panel A of Fig 2, the results generally meet the expectation of unimodality, but there are many deviations in the distribution. When we examine the distri.