Practice of culture history in archaeology, we treat unimodality as a construct that serves with continuity to help identify patterns that are potentially the product of cultural transmission. Second, generation of candidate solutions should be automated, so that seriation can be used as part of larger analyses (e.g.,PLOS ONE | DOI:10.1371/journal.pone.0124942 April 29,8 /The IDSS Frequency Seriation AlgorithmFig 2. The results of a probabilistic seriation analysis for a set of late prehistoric ceramics assemblages from the (S)-(-)-Blebbistatin manufacturer Memphis and St. Francis areas as described by Lipo [84] and Phillips and colleagues [10]. Here, the figures show the the results of correspondence analysis (CA) for the dataset in Table 2 MK-5172 web following [85]. (A) The seriation order produced from the CA shown in standard centered bar format. (B) CA results shown with clusters as determined by hierarchical cluster analysis on the principlePLOS ONE | DOI:10.1371/journal.pone.0124942 April 29,9 /The IDSS Frequency Seriation Algorithmcomponents. One can see that the change in the frequencies of types roughly follows a unimodal distribution, but there are numerous violations of unimodality as well. Data and R code for the correspondence analysis are available at https://github.com/mmadsen/lipomadsen2015-idss-seriation-paper. doi:10.1371/journal.pone.0124942.gspatial analysis, simulation studies of migration, trade, or cultural transmission). Third, the algorithm should provide error estimates and confidence bands where possible, to allow evaluation of the quality of a solution given the input data, and diagnosis of any violations of unimodality or continuity. Finally, the technique must be able to find all viable deterministic solutions given bounded and reasonable processing time for even relatively large sets of assemblage (e.g., 20 or 50), such that resampling or the bootstrap can be used to calculate error terms and evaluate the effects of sample size. These are not easy requirements to meet. In the space created by all the possible orderings of assemblages, the vast majority of orders are invalid, as the combinations violate the conditions of the DFS method due to deviations from unimodality and/or continuity. Even with stricter constraints on possible solutions, valid candidates cannot be found by enumeration for more than a handful of assemblages. The search space must be “pruned” in some fashion to remove combinations that cannot possibly be part of a valid solution. Overview of the IDSS Algorithm. The technique we propose to accomplish these goals is called the Iterative Deterministic Seriation Solution (IDSS). IDSS builds DFS orders in an iterative process, starting with valid seriation solutions composed of the smallest possible number of assemblages and then employing these as building blocks for larger solutions. Solutions are grown from valid smaller solutions instead of enumerating possible combinations. We start with combinations of three assemblages (triples), the fewest number that can be evaluated in terms of the degree to which they meet the demands of the model. With three assemblages, we retain only those sets in which the frequencies for each of the classes show a steady increase, steady decrease, a middle “peak,” or no change at all (Fig 4). Assemblage orders that have frequencies that decrease then increase are eliminated as building blocks. The next step in the procedure is to use just the successful triples and see if any of the remaining assemblages.Practice of culture history in archaeology, we treat unimodality as a construct that serves with continuity to help identify patterns that are potentially the product of cultural transmission. Second, generation of candidate solutions should be automated, so that seriation can be used as part of larger analyses (e.g.,PLOS ONE | DOI:10.1371/journal.pone.0124942 April 29,8 /The IDSS Frequency Seriation AlgorithmFig 2. The results of a probabilistic seriation analysis for a set of late prehistoric ceramics assemblages from the Memphis and St. Francis areas as described by Lipo [84] and Phillips and colleagues [10]. Here, the figures show the the results of correspondence analysis (CA) for the dataset in Table 2 following [85]. (A) The seriation order produced from the CA shown in standard centered bar format. (B) CA results shown with clusters as determined by hierarchical cluster analysis on the principlePLOS ONE | DOI:10.1371/journal.pone.0124942 April 29,9 /The IDSS Frequency Seriation Algorithmcomponents. One can see that the change in the frequencies of types roughly follows a unimodal distribution, but there are numerous violations of unimodality as well. Data and R code for the correspondence analysis are available at https://github.com/mmadsen/lipomadsen2015-idss-seriation-paper. doi:10.1371/journal.pone.0124942.gspatial analysis, simulation studies of migration, trade, or cultural transmission). Third, the algorithm should provide error estimates and confidence bands where possible, to allow evaluation of the quality of a solution given the input data, and diagnosis of any violations of unimodality or continuity. Finally, the technique must be able to find all viable deterministic solutions given bounded and reasonable processing time for even relatively large sets of assemblage (e.g., 20 or 50), such that resampling or the bootstrap can be used to calculate error terms and evaluate the effects of sample size. These are not easy requirements to meet. In the space created by all the possible orderings of assemblages, the vast majority of orders are invalid, as the combinations violate the conditions of the DFS method due to deviations from unimodality and/or continuity. Even with stricter constraints on possible solutions, valid candidates cannot be found by enumeration for more than a handful of assemblages. The search space must be “pruned” in some fashion to remove combinations that cannot possibly be part of a valid solution. Overview of the IDSS Algorithm. The technique we propose to accomplish these goals is called the Iterative Deterministic Seriation Solution (IDSS). IDSS builds DFS orders in an iterative process, starting with valid seriation solutions composed of the smallest possible number of assemblages and then employing these as building blocks for larger solutions. Solutions are grown from valid smaller solutions instead of enumerating possible combinations. We start with combinations of three assemblages (triples), the fewest number that can be evaluated in terms of the degree to which they meet the demands of the model. With three assemblages, we retain only those sets in which the frequencies for each of the classes show a steady increase, steady decrease, a middle “peak,” or no change at all (Fig 4). Assemblage orders that have frequencies that decrease then increase are eliminated as building blocks. The next step in the procedure is to use just the successful triples and see if any of the remaining assemblages.