Optimization for scalable analytics using time series models
Techniques are described for optimizing scalability of analytics that use time-series models. In one or more embodiments, a stored time-series model includes a plurality of data points representing seasonal behavior in a training set of time-series data for at least one season. A target time for evaluating the time-series model is then determined, and the target time or one or more times relative to the target time are mapped to a subset of the plurality of data points. Based on the mapping, a trimmed version of the time-series model is generated by loading the subset of the plurality of data points into a cache, the subset of data points representing seasonal behavior in the training set of time-series data for a portion of the at least one season. A target set of time-series data may be evaluated suing the trimmed version of the time-series in the cache.
1. A method comprising:
storing, by one or more hardware processors, a time-series model, the time-series model including a plurality of data points representing seasonal behavior in a training set of time-series data for at least one season, wherein the plurality of data points do not include an explicit timestamp;
determining, by the one or more hardware processors, a target time for evaluating the time-series model;
mapping, by the one or more hardware processors to a subset of the plurality of data points, the target time or one or more times relative to the target time;
based on the mapping, generating, by the one or more hardware processors, a trimmed version of the time-series model in a cache by loading the subset of the plurality of data points into the cache, the subset of data points representing seasonal behavior in the training set of time-series data for a portion of the at least one season within a threshold range of the target time;
determining, by the one or more hardware processors, whether a target set of time-series data includes data points outside the threshold range of the target time;
responsive to determining that the target set of time-series data includes data points outside the threshold range of the target time, updating, by the one or more hardware processors, the trimmed version of the time-series model in the cache by loading a second subset of the plurality of data points into the cache; and
evaluating, by the one or more hardware processors, the target set of time-series data using the updated trimmed version of the time-series model in the cache to output an evaluation result.
2. The method of claim 1 , wherein the target time is determined based on at least one of a current time or a time associated with the target set of time-series data.
3. The method of claim 1 , further comprising determining a new target time for evaluating the time-series model; and responsive to determining the new target time, updating the trimmed version of the time-series model in the cache by loading a third subset of the plurality of data points into the cache, the third subset of data points representing seasonal behavior in the training set of time-series data for a second portion of the at least one season.
4. The method of claim 1 , wherein the mapping is based on a variance in the data points representing seasonal behavior in the training set of time-series data.
5. The method of claim 1 , wherein the trimmed version of the time-series model does not include data points that do not represent seasonal behavior.
6. The method of claim 1 , further comprising retraining the trimmed version of the time-series model using the target set of time-series data.
7. The method of claim 1 , further comprising: detecting a change in the time-series model, the change causing the time-series model to include data points representing seasonal behavior for a different season; responsive to detecting the change in the time-series model, updating the trimmed version of the time-series model to include a greater number or lesser number of data points.
8. The method of claim 7 wherein the change in the time-series model is from a daily season to a weekly season.
9. One or more non-transitory computer-readable media storing instructions which, when executed by one or more hardware processors, cause performance of operations comprising:
storing a time-series model, the time-series model including a plurality of data points representing seasonal behavior in a training set of time-series data for at least one season, wherein the plurality of data points do not include an explicit timestamp;
determining a target time for evaluating the time-series model;
mapping, to a subset of the plurality of data points, the target time or one or more times relative to the target time;
based on the mapping, generating a trimmed version of the time-series model in a cache by loading the subset of the plurality of data points into the cache, the subset of data points representing seasonal behavior in the training set of time-series data for a portion of the at least one season within a threshold range of the target time;
determining whether a target set of time-series data includes data points outside the threshold range of the target time;
responsive to determining that the target set of time-series data includes data points outside the threshold range of the target time, updating the trimmed version of the time-series model in the cache by loading a second subset of the plurality of data points into the cache; and
evaluating the target set of time-series data using the updated trimmed version of the time-series model in the cache to output an evaluation result.
10. The one or more non-transitory computer-readable media of claim 9 , wherein the target time is determined based on at least one of a current time or a time associated with the target set of time-series data.
11. The one or more non-transitory computer-readable media of claim 9 , wherein the instructions further cause operations comprising determining a new target time for evaluating the time-series model; and responsive to determining the new target time, updating the trimmed version of the time-series model in the cache by loading a third subset of the plurality of data points into the cache, the third subset of data points representing seasonal behavior in the training set of time-series data for a second portion of the at least one season.
12. The one or more non-transitory computer-readable media of claim 9 , wherein the mapping is based on a variance in the data points representing seasonal behavior in the training set of time-series data.
13. The one or more non-transitory computer-readable media of claim 9 , wherein the trimmed version of the time-series model does not include data points that do not represent seasonal behavior.
14. The one or more non-transitory computer-readable media of claim 9 , the instructions further causing operations comprising retraining the trimmed version of the time-series model using the target set of time-series data.
15. The one or more non-transitory computer-readable media of claim 9 , the instructions further causing operations comprising: detecting a change in the time-series model, the change causing the time-series model to include data points representing seasonal behavior for a different season; responsive to detecting the change in the time-series model, updating the trimmed version of the time-series model to include a greater number or lesser number of data points.
16. The one or more non-transitory computer-readable media of claim 15 wherein the change in the time-series model is from a daily season to a weekly season.
17. A system comprising:
one or more hardware processors;
one or more non-transitory computer-readable media storing instructions which, when executed by the one or more hardware processors, cause:
storing a time-series model, the time-series model including a plurality of data points representing seasonal behavior in a training set of time-series data for at least one season, wherein the plurality of data points do not include an explicit timestamp;
determining a target time for evaluating the time-series model;
mapping, to a subset of the plurality of data points, the target time or one or more times relative to the target time;
based on the mapping, generating a trimmed version of the time-series model in a cache by loading the subset of the plurality of data points into the cache, the subset of data points representing seasonal behavior in the training set of time-series data for a portion of the at least one season within a threshold range of the target time;
determining whether a target set of time-series data includes data points outside the threshold range of the target time;
responsive to determining that the target set of time-series data includes data points outside the threshold range of the target time, updating the trimmed version of the time-series model in the cache by loading a second subset of the plurality of data points into the cache; and
evaluating the target set of time-series data using the updated trimmed version of the time-series model in the cache to output an evaluation result.
18. The system of claim 17 , wherein the target time is determined based on at least one of a current time or a time associated with the target set of time-series data.
19. The system of claim 17 , wherein the instructions further cause: determining a new target time for evaluating the time-series model; and responsive to determining the new target time, updating the trimmed version of the time-series model in the cache by loading a third subset of the plurality of data points into the cache, the third subset of data points representing seasonal behavior in the training set of time-series data for a second portion of the at least one season.
20. The system of claim 17 , wherein the instructions further cause: detecting a change in the time-series model, the change causing the time-series model to include data points representing seasonal behavior for a different season; responsive to detecting the change in the time-series model, updating the trimmed version of the time-series model to include a greater number or lesser number of data points.