SEASONALITY VALIDATION AND DETERMINATION OF PATTERNS

• US 20170364851A1
• Filed: 06/20/2016
• Published: 12/21/2017
• Est. Priority Date: 06/20/2016
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: Maheshwari, Arvind Kumar, Ravichandran, Raghav, Volchegursky, Vladimir, Huang, Tse-Han
• Granted Patent: US 11,263,566 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 8/65 Updates security arrangements therefor G06F21/57
• CPC Class Codes: G06F 9/50 Allocation of resources, e.g. of the central processing unit [CPU]
• CPC Class Codes: G06Q 10/06314 Calendaring for a resource
• CPC Class Codes: G06Q 10/06315 Needs-based resource requirements planning or analysis

Abstract

Techniques are described herein for seasonal pattern determination and validation. In one or more embodiments, a set of time-series data is received to analyze for seasonal behavior. In response a plurality of patterns may be generated, including a first pattern and a second pattern, such that each of the first pattern and the second pattern approximate data points that represent a same sub-period of multiple instances of a season within the set of time-series data. One or more other instances of the season may then be analyzed to determine whether at least part of the first pattern or the second pattern is detected. Based at least in part on determining that the at least part of the first pattern is detected in the at least part of the same sub-period, a responsive action that is associated with the first pattern may be performed.

Claims

1. A method comprising:

receiving, by one or more computing devices, a set of time-series data;

generating, by one or more computing devices, a first pattern and a second pattern such that;

each of the first pattern and the second pattern approximate data points of a same sub-period of multiple instances of a season within the set of time-series data;

in the set of time-series data, a first set of instances of the season contain corresponding data points that align more closely to the first pattern than to at least the second pattern;

in the set of time-series data, a second set of instances of the season contain corresponding data points that align more closely to the second pattern than to at least the first pattern; and

the first pattern is different from the second pattern;

analyzing at least part of the same sub-period of one or more other instances of the season to determine whether at least part of the first pattern is detected in the at least part of the same sub-period of the one or more other instances of the season;

based at least in part on determining that the at least part of the first pattern is detected in the at least part of the same sub-period of the one or more other instances of the season, performing a responsive action associated with the first pattern.

2. The method of claim 1, wherein a sequence of data points align more closely to the first pattern than to the at least the second pattern if a similarity measure indicates that the sequence of data points is more similar to the first pattern than the second pattern.

3. The method of claim 2, wherein the similarity measure is determined based, at least in part, on a cosine similarity function that compares a vector including the sequence of data points with a vector including data points approximated by the first pattern.

4. The method of claim 1 wherein generating, by one or more computing devices, the first pattern and the second pattern comprises:

generating, by the one or more computing devices within volatile or non-volatile storage, a first cluster that groups the first set of instances and a second cluster that groups the second set of instances;

computing a first centroid of the first cluster and a second centroid of the second cluster;

wherein the first pattern is generated based, at least in part, on the first centroid of the first cluster;

wherein the second pattern is generated based, at least in part, on the second centroid of the second cluster.

5. The method of claim 1, wherein determining that the at least part of the first pattern is detected in the at least part of the same sub-period of the one or more other instances of the season comprises determining that the at least part of the same sub-period aligns more closely with the at least part of the first pattern or at least part of the second pattern.

6. The method of claim 1, further comprising identifying the sub-period for which to generate the first pattern based, at least in part, on a determination of which data points within the first set of instances weaken a strength of seasonality between the first set of instances.

7. The method of claim 1, further comprising computing a strength of seasonality for the first pattern based, at least in part, on a similarity measure that indicates how similar the same sub-period is within the first set of instances of the season;

wherein the responsive action associated with the first pattern is performed, based at least in part, on the seasonality strength for the first pattern.

8. The method of claim 1, wherein the season corresponds to a week of time and the same sub-period corresponds to one or more days within the week.

9. The method of claim 1, wherein the responsive action includes one or more of performing a capacity planning operation based on the first pattern or generating a notification based on the first pattern.

10. A method comprising:

receiving, by one or more computing devices, a set of time-series data;

determining, by one or more computing devices, a first pattern that approximates data points from an at least first sub-period of a set of instances of a season within the set of time-series data;

determining, by one or more computing devices, a second pattern that approximates data points from an at least second sub-period of the set of instances of the season within the set of time-series data;

wherein the second pattern is different than the first pattern and the at least second sub-period is different than the at least first sub-period;

determining that a first strength of seasonality associated with the first pattern is weaker than a second strength of seasonality associated with the second pattern;

in response to determining that the first strength of seasonality associated with the first pattern is weaker than the second strength of seasonality associated with the second pattern, generating an analytic output that is different than if the first strength of seasonality associated with the first pattern was stronger than the second strength of seasonality associated with the second pattern.

11. The method of claim 10, wherein determining, by one or more computing devices, a second pattern that approximates data points from an at least second sub-period of the set of instances of the season within the set of time-series data comprises:

identifying a subset of data points approximated by the first pattern that weaken the first strength of seasonality of the first pattern;

generating the second pattern by truncating the subset of data points from the first pattern.

12. The method of claim 10, wherein the at least first sub-period spans either a whole or a part of the season;

wherein the at least second sub-period spans part of the at least first sub-period.

13. The method of claim 10, further comprising:

receiving an input that indicates a first set of one or more windows of time within the season for which to generate a seasonal pattern;

wherein the at least first sub-period corresponds to the first set of one or more windows of time;

wherein the at least second sub-period corresponds to a second set of one or more windows of time that is different than the first set of one or more windows of time;

wherein the analytic output indicates that seasonality is stronger for the second set of one or more windows of time than for the first set of one or more windows of time.

14. The method of claim 10, wherein the analytic output ranks a plurality of patterns including the first pattern and the second pattern based on strength of seasonality.

15. The method of claim 14, further comprising:

determining, by one or more computing devices, a third pattern that approximates data points from at least a third sub-period of a second set of instances from a second season;

wherein the second season has a different duration than the first season;

wherein the second set of instances include at least some of the data points from the first set of instances of the first season;

wherein the plurality of patterns includes the third pattern.

16. One or more non-transitory computer-readable media storing instructions which, when executed by one or more computing devices, cause operations comprising:

receiving a set of time-series data;

generating a first pattern and a second pattern such that;

each of the first pattern and the second pattern approximate data points of a same sub-period of multiple instances of a season within the set of time-series data;

in the set of time-series data, a first set of instances of the season contain corresponding data points that align more closely to the first pattern than to at least the second pattern;

in the set of time-series data, a second set of instances of the season contain corresponding data points that align more closely to the second pattern than to at least the first pattern; and

the first pattern is different from the second pattern;

analyzing at least part of the same sub-period of one or more other instances of the season to determine whether at least part of the first pattern is detected in the at least part of the same sub-period of the one or more other instances of the season;

based at least in part on determining that the at least part of the first pattern is detected in the at least part of the same sub-period of the one or more other instances of the season, performing a responsive action associated with the first pattern.

17. The one or more non-transitory computer-readable media of claim 16, wherein a sequence of data points align more closely to the first pattern than to the at least the second pattern if a similarity measure indicates that the sequence of data points is more similar to the first pattern than the second pattern.

18. The one or more non-transitory computer-readable media of claim 17, wherein the similarity measure is determined based, at least in part, on a cosine similarity function that compares a vector including the sequence of data points with a vector including data points approximated by the first pattern.

19. The one or more non-transitory computer-readable media of claim 16, wherein instructions for generating the first pattern and the second pattern comprise instructions, which when executed by one or more computing devices, cause operations comprising:

generating, within volatile or non-volatile storage, a first cluster that groups the first set of instances and a second cluster that groups the second set of instances;

computing a first centroid of the first cluster and a second centroid of the second cluster;

wherein the first pattern is generated based, at least in part, on the first centroid of the first cluster;

wherein the second pattern is generated based, at least in part, on the second centroid of the second cluster.

20. The one or more non-transitory computer-readable media of claim 16, wherein instructions for determining that the at least part of the first pattern is detected in the at least part of the same sub-period of the one or more other instances of the season comprise instructions, which when executed by one or more computing devices, cause determining that the at least part of the same sub-period aligns more closely with the at least part of the first pattern or at least part of the second pattern.

21. The one or more non-transitory computer-readable media of claim 16, further storing instruction, which when executed by one or more computing devices, cause operations comprising identifying the sub-period for which to generate the first pattern based, at least in part, on a determination of which data points within the first set of instances weaken a strength of seasonality between the first set of instances.

22. The one or more non-transitory computer-readable media of claim 16, further storing instruction, which when executed by one or more computing devices, cause operations comprising computing a strength of seasonality for the first pattern based, at least in part, on a similarity measure that indicates how similar the same sub-period is within the first set of instances of the season;

wherein the responsive action associated with the first pattern is performed, based at least in part, on the seasonality strength for the first pattern.

23. The one or more non-transitory computer-readable media of claim 16, wherein the season corresponds to a week of time and the same sub-period corresponds to one or more days within the week.

24. The one or more non-transitory computer-readable media of claim 16, wherein the responsive action includes one or more of performing a capacity planning operation based on the first pattern or generating a notification based on the first pattern.

25. One or more non-transitory computer-readable media storing instructions which, when executed by one or more computing devices, cause operations comprising:

receiving a set of time-series data;

determining a first pattern that approximates data points from an at least first sub-period of a set of instances of a season within the set of time-series data;

determining a second pattern that approximates data points from an at least second sub-period of the set of instances of the season within the set of time-series data;

wherein the second pattern is different than the first pattern and the at least second sub-period is different than the at least first sub-period;

determining that a first strength of seasonality associated with the first pattern is weaker than a second strength of seasonality associated with the second pattern;

in response to determining that the first strength of seasonality associated with the first pattern is weaker than the second strength of seasonality associated with the second pattern, generating an analytic output that is different than if the first strength of seasonality associated with the first pattern was stronger than the second strength of seasonality associated with the second pattern.

26. The one or more non-transitory computer-readable media of claim 25, wherein instructions for determining, a second pattern that approximates data points from an at least second sub-period of the set of instances of the season within the set of time-series data comprise instructions, which when executed by one or more computing devices, cause operations comprising:

identifying a subset of data points approximated by the first pattern that weaken the first strength of seasonality of the first pattern;

generating the second pattern by truncating the subset of data points from the first pattern.

27. The one or more non-transitory computer-readable media of claim 25, wherein the at least first sub-period spans either a whole or a part of the season;

wherein the at least second sub-period spans part of the at least first sub-period.

28. The one or more non-transitory computer-readable media of claim 25, further storing instructions, which when executed by one or more computing devices, cause operations comprising:

receiving an input that indicates a first set of one or more windows of time within the season for which to generate a seasonal pattern;

wherein the at least first sub-period corresponds to the first set of one or more windows of time;

wherein the at least second sub-period corresponds to a second set of one or more windows of time that is different than the first set of one or more windows of time;

wherein the analytic output indicates that seasonality is stronger for the second set of one or more windows of time than for the first set of one or more windows of time.

29. The one or more non-transitory computer-readable media of claim 25, wherein the analytic output ranks a plurality of patterns including the first pattern and the second pattern based on strength of seasonality.

30. The one or more non-transitory computer-readable media of claim 29, further storing instructions, which when executed by one or more computing devices, cause operations comprising:

determining, by one or more computing devices, a third pattern that approximates data points from at least a third sub-period of a second set of instances from a second season;

wherein the second season has a different duration than the first season;

wherein the second set of instances include at least some of the data points from the first set of instances of the first season;

wherein the plurality of patterns includes the third pattern.