Computer implemented systems for automatic hierarchy for large scale time series data sets

A computer-implemented method, comprising:

• receiving timestamped data, wherein the timestamped data is associated with time-independent attributes and one or more independent variables;
• clustering the timestamped data into a set of clusters, wherein clustering the timestamped data comprises:
• detecting patterns in the timestamped data,
• determining how the timestamped data is influenced by the set of one or more independent variables, and
• generating the set of clusters based on the patterns of the timestamped data and how the timestamped data is influenced by the one or more independent variables;
• selecting attributes from among the time-independent attributes based on the set of clusters, wherein the selected attributes are a subset of the time-independent attributes;
• ordering the selected attributes, wherein ordering the selected attributes comprises sequentially building an order of the selected attributes or globally building the order of the selected attributes;
• wherein sequentially building the order of the selected attributes comprises sequentially determining an order position that a given attribute of the selected attributes is to have in a current order of the selected attributes, updating the current order of the selected attributes based on the order position for the given attribute, and determining an order position that a next attribute of the selected attributes is to have in the updated current order;
• and wherein globally building the order of the selected attributes comprises determining a first order of the selected attributes, determining a second order of the selected attributes, and comparing the first order and the second order to select the first order or the second order;
• generating a structure for a hierarchy using the order of the selected attributes, wherein the structure for the hierarchy defines a plurality of levels of the hierarchy;
• and
• generating predicted values across the plurality of levels of the hierarchy using the received timestamped data and the generated hierarchical structure, wherein predicted values of one or more levels of the plurality of levels is informed by predicted values of another level of the plurality of levels.

A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause a data processing apparatus to perform operations including:

• receiving timestamped data, wherein the timestamped data is associated with a time-independent attributes and one or more independent variables;
• clustering the timestamped data into a set of clusters, wherein clustering the timestamped data comprises:
• detecting patterns in the timestamped data,
• determining how the timestamped data is influenced by the one or more independent variables, and
• generating the set of clusters based on the patterns of the timestamped data and how the timestamped data is influenced by the one or more independent variables;
• selecting attributes from among the time-independent attributes based on the set of clusters, wherein the selected attributes are a subset of the time-independent attributes;
• ordering the selected attributes, wherein ordering the selected attributes comprises sequentially building an order of the selected attributes or globally building the order of the selected attributes;
• wherein sequentially building the order of the selected attributes comprises sequentially determining an order position that a given attribute of the selected attributes is to have in a current order of the selected attributes, updating the current order of the selected attributes based on the order position for the given attribute, and determining an order position that a next attribute of the selected attributes is to have in the updated current order;
• and wherein globally building the order of the selected attributes comprises determining a first order of the selected attributes, determining a second order of the selected attributes, and comparing the first order and the second order to select the first order or the second order;
• generating a structure for a hierarchy using the order of the selected attributes, wherein the structure for the hierarchy defines a plurality of levels of the hierarchy;
• and
• generating predicted values across the plurality of levels of the hierarchy using the received timestamped data and the generated hierarchical structure, wherein predicted values of one or more levels of the plurality of levels is informed by predicted values of another level of the plurality of levels.