Abstract: Techniques are described for automated selection of components for a generalized linear model. In one example, a method includes determining a candidate set of distributions, a candidate set of link functions, and a candidate set of predictor variables, based at least in part on a dataset of interest. The method further includes selecting a distribution from the initial candidate set of distributions and a link function from the initial candidate set of link functions, based at least in part on the candidate set of predictor variables; and selecting predictor variables from the candidate set of predictor variables, based at least in part on the selected distribution and the selected link function. The method further includes reiterating the selecting processes until a stopping criterion is fulfilled, and generating a generalized linear model output comprising the selected distribution, the selected link function, and the selected predictor variables.

Abstract: Techniques are described for generating characterizations of time series data. In one example, a method includes extracting a trend-cycle component, a seasonal component, and an irregular component from a time series of data. The method further includes performing one or more pattern analyses on the trend-cycle component, the seasonal component, and the irregular component. The method further includes, for each pattern analysis of the one or more pattern analyses, performing a comparison of an analytic result of the respective pattern analysis to a selected significance threshold for the respective pattern analysis to determine if the analytic result passes the significance threshold for the respective pattern analysis. The method further includes generating an output for each of the analytic results that pass the significance threshold for the respective pattern analysis.

Abstract: Techniques are described for generating characterizations of time series data. In one example, a method includes extracting a trend-cycle component, a seasonal component, and an irregular component from a time series of data. The method further includes performing one or more pattern analyses on the trend-cycle component, the seasonal component, and the irregular component. The method further includes, for each pattern analysis of the one or more pattern analyses, performing a comparison of an analytic result of the respective pattern analysis to a selected significance threshold for the respective pattern analysis to determine if the analytic result passes the significance threshold for the respective pattern analysis. The method further includes generating an output for each of the analytic results that pass the significance threshold for the respective pattern analysis.

Abstract: One or more processors generate subsets of cluster feature (CF)-trees, which represent respective sets of local data as leaf entries. One or more processors collect variables that were used to generate the CF-trees included in the subsets. One or more processors generate respective approximate clustering solutions for the subsets by applying hierarchical agglomerative clustering to the collected variables and leaf entries of the plurality of CF-trees. One or more processors select candidate sets of variables with maximal goodness that are locally optimal for respective subsets based on the approximate clustering solutions. One or more processors select a set of variables, which produce an overall clustering solution, from the candidate sets of variables.

Abstract: A computing device includes at least one processor, and at least one module operable by the at least one processor to receive data representing a hierarchy, wherein the hierarchy comprises at least one set of sibling nodes and a respective parent node, generate a condensed hierarchy by determining a grouping for the at least one set of sibling nodes, determine whether the at least one set of sibling nodes can be represented by the respective parent node, based at least in part on the grouping for the at least one set of sibling nodes, and responsive to determining that the at least one set of sibling nodes can be represented by the respective parent node, remove the at least one set of sibling nodes from the condensed hierarchy. The at least one module may further be operable by the at least one processor to output the condensed hierarchy for display.

Abstract: One or more processors generate subsets of cluster feature (CF)-trees, which represent respective sets of local data as leaf entries. One or more processors collect variables that were used to generate the CF-trees included in the subsets. One or more processors generate respective approximate clustering solutions for the subsets by applying hierarchical agglomerative clustering to the collected variables and leaf entries of the plurality of CF-trees. One or more processors select candidate sets of variables with maximal goodness that are locally optimal for respective subsets based on the approximate clustering solutions. One or more processors select a set of variables, which produce an overall clustering solution, from the candidate sets of variables.

Abstract: Provided are techniques for imputing a missing value for each of one or more predictor variables. Data is received from one or more data sources. For each of the one or more predictor variables, an imputation model is built based on information of a target variable; a type of imputation model to construct is determined based on the one or more data sources, a measurement level of the predictor variable, and a measurement level of the target variable; and the determined type of imputation model is constructed using basic statistics of the predictor variable and the target variable. The missing value is imputed for each of the one or more predictor variables using the data from the one or more data sources and one or more built imputation models to generate a completed data set.

Abstract: A method for comparing predictive data models based on a predictive model search is provided. The method may include receiving a first and second portion of a set of data. The method may also include identifying a first and second variation of the second portion, wherein the first variation is different from the second variation. The method may further include generating first predictive data models based on the first variation, and second predictive data models based on the second variation. Additionally, the method may include applying a criteria to rank the first and second predictive data models based on predictive strength. The method may also include presenting a display of the ranked criteria, comprising the first portion, and a portion of the first and second predictive data models, wherein the portion of the first and second predictive data models are collectively ranked and presented according to the predictive strength.

Abstract: A method for comparing predictive data models based on a predictive model search is provided. The method may include receiving a first and second portion of a set of data. The method may also include identifying a first and second variation of the second portion, wherein the first variation is different from the second variation. The method may further include generating first predictive data models based on the first variation, and second predictive data models based on the second variation. Additionally, the method may include applying a criteria to rank the first and second predictive data models based on predictive strength. The method may also include presenting a display of the ranked criteria, comprising the first portion, and a portion of the first and second predictive data models, wherein the portion of the first and second predictive data models are collectively ranked and presented according to the predictive strength.

Abstract: Provided are techniques for interaction detection for generalized linear models. Basic statistics are calculated for a pair of categorical predictor variables and a target variable from a dataset during a single pass over the dataset. It is determined whether there is a significant interaction effect for the pair of categorical predictor variables on the target variable by: calculating a log-likelihood value for a full generalized linear model without estimating model parameters; calculating the model parameters for a reduced generalized linear model with a recursive marginal mean accumulation technique using the basic statistics; calculating a log-likelihood value for the reduced generalized linear model; calculating a likelihood ratio test statistic using the log-likelihood value for the full generalized linear model and the log-likelihood value for the reduced generalized linear model; calculating a p-value of the likelihood ratio test statistic; and comparing the p-value to a significance level.

Abstract: Techniques for presenting insight into classification trees may include performing a grouping analysis to group leaf nodes of a classification tree into a significant group and an insignificant group, performing influential target category analysis to identify one or more influential target categories for the leaf nodes of the classification tree in the significant group, and presenting one or more insights into the classification tree based on the grouping analysis and the influential target category analysis. Techniques for presenting insight into regression trees may include performing a grouping analysis to group leaf nodes of a regression tree into a high group and a low group, performing unusual node detection analysis to detect one or more outlier nodes in the high group and in the low group, and presenting one or more insights into the regression tree based on the grouping analysis and the unusual node detection analysis.

Abstract: Techniques for presenting insight into classification trees may include performing a grouping analysis to group leaf nodes of a classification tree into a significant group and an insignificant group, performing influential target category analysis to identify one or more influential target categories for the leaf nodes of the classification tree in the significant group, and presenting one or more insights into the classification tree based on the grouping analysis and the influential target category analysis. Techniques for presenting insight into regression trees may include performing a grouping analysis to group leaf nodes of a regression tree into a high group and a low group, performing unusual node detection analysis to detect one or more outlier nodes in the high group and in the low group, and presenting one or more insights into the regression tree based on the grouping analysis and the unusual node detection analysis.

Abstract: In one example, a method includes determining, based on historical purchase data for a customer, an expectancy value that indicates when the customer is expected to make a purchase from a business, determining, based on the historical purchase data for the customer, a frequency value that indicates at what frequency the customer is expected to make purchases from the business during a future time period, and determining, based on the historical purchase data for the customer, a monetary value that indicates how much the customer is expected to spend during the future time period. In this example, the method includes determining, based on the expectancy value, the frequency value, and the monetary value, a future customer value score that indicates how valuable the customer is likely to be in the future time period.

Abstract: In one example, a method includes determining, based on historical purchase data for a customer, an expectancy value that indicates when the customer is expected to make a purchase from a business, determining, based on the historical purchase data for the customer, a frequency value that indicates at what frequency the customer is expected to make purchases from the business during a future time period, and determining, based on the historical purchase data for the customer, a monetary value that indicates how much the customer is expected to spend during the future time period. In this example, the method includes determining, based on the expectancy value, the frequency value, and the monetary value, a future customer value score that indicates how valuable the customer is likely to be in the future time period.

Abstract: Provided are techniques for computing a task result. A processing data set of records is created, wherein each of the records contains data specific to a sub-task from a set of actual sub-tasks and contains a reference to data shared by the set of actual sub-tasks, and wherein a number of the records is equivalent to a number of the actual sub-tasks in the set of actual sub-tasks. With each mapper in a set of mappers, one of the records of the processing data set is received and an assigned sub-task is executed using the received one of the records to generate output. With a single reducer, the output from each mapper in the set of mappers is reduced to determine a task result.

Abstract: One or more processors generate subsets of cluster feature (CF)-trees, which represent respective sets of local data as leaf entries. One or more processors collect variables that were used to generate the CF-trees included in the subsets. One or more processors generate respective approximate clustering solutions for the subsets by applying hierarchical agglomerative clustering to the collected variables and leaf entries of the plurality of CF-trees. One or more processors select candidate sets of variables with maximal goodness that are locally optimal for respective subsets based on the approximate clustering solutions. One or more processors select a set of variables, which produce an overall clustering solution, from the candidate sets of variables.

Abstract: One or more processors initiate cluster feature (CF)-tree based hierarchical clustering on leaf entries of CF-trees included in a plurality of subsets. One or more processors, generate respective partial clustering solutions for the subsets. A partial clustering solution includes a set of regular sub-clusters and candidate outlier sub-clusters. One or more processors generate initial regular clusters by performing hierarchical clustering using the regular sub-clusters. For a candidate outlier sub-cluster, one or more processors determine a closest initial regular cluster, and a distance separating the candidate outlier sub-cluster and the closest initial regular cluster. One or more processors determine which candidate outlier sub-clusters are outlier clusters based on which candidate outlier sub-clusters have a computed distance to their respective closest initial regular cluster that is greater than a corresponding distance threshold associated with their respective closest initial regular cluster.

Abstract: One or more processors generate subsets of cluster feature (CF)-trees, which represent respective sets of local data as leaf entries. One or more processors collect variables that were used to generate the CF-trees included in the subsets. One or more processors generate respective approximate clustering solutions for the subsets by applying hierarchical agglomerative clustering to the collected variables and leaf entries of the plurality of CF-trees. One or more processors select candidate sets of variables with maximal goodness that are locally optimal for respective subsets based on the approximate clustering solutions. One or more processors select a set of variables, which produce an overall clustering solution, from the candidate sets of variables.

Abstract: One or more processors initiate cluster feature (CF)-tree based hierarchical clustering on leaf entries of CF-trees included in a plurality of subsets. One or more processors, generate respective partial clustering solutions for the subsets. A partial clustering solution includes a set of regular sub-clusters and candidate outlier sub-clusters. One or more processors generate initial regular clusters by performing hierarchical clustering using the regular sub-clusters. For a candidate outlier sub-cluster, one or more processors determine a closest initial regular cluster, and a distance separating the candidate outlier sub-cluster and the closest initial regular cluster. One or more processors determine which candidate outlier sub-clusters are outlier clusters based on which candidate outlier sub-clusters have a computed distance to their respective closest initial regular cluster that is greater than a corresponding distance threshold associated with their respective closest initial regular cluster.

Abstract: Provided are techniques for computing a task result. A processing data set of records is created, wherein each of the records contains data specific to a sub-task from a set of actual sub-tasks and contains a reference to data shared by the set of actual sub-tasks, and wherein a number of the records is equivalent to a number of the actual sub-tasks in the set of actual sub-tasks. With each mapper in a set of mappers, one of the records of the processing data set is received and an assigned sub-task is executed using the received one of the records to generate output. With a single reducer, the output from each mapper in the set of mappers is reduced to determine a task result.