Hellinger Distance for ALEC

Description

One of the main issues we have with the way we are currently trying to group alarms into situations is the usage of timestamps.

Both our engines (DBSCAN and Deep Learning), use the offset between the ending time the two alarms is trying to associate. This doesn't take into account the alarm duration.

By using Hellinger Distance we attempt to fit these start and end timestamps of alarms into a distribution which would allow us to calculate a more robust distance. Hellinger Distance between alarms would be a measure between 0 (identical alarms in terms of time) and 1 (alarms have no time overlap).

Our hypothesis is that this measure can improve the situation generation process for either engine.

Goals:

  • Learn how to calculate Hellinger distance. (ALEC-102)

  • Implement Hellinger distance for machine learning engines:

    • Clustering (DBSCAN) (ALEC-122).

    • Deep Learning (ALEC-XXX).

  • Assess if Hellinger Distance has a positive impact in ALEC: (ALEC-107 and ALEC-XXX)

Target:

  • Network Operator / Administrator, looking at our alarm dashboard.

If we can provide better accuracy, we will let the Network Operator / Administrator be more efficient at reducing the Mean Time to Repair (by having a more focused approach).

Acceptance / Success Criteria

None
66% Done
Loading...

Lucidchart Diagrams

Activity

Show:
Fixed

Details

Assignee

Reporter

Labels

Original estimate

Time tracking

No time logged1d remaining

Sprint

Fix versions

Due date

Priority

Epic Name

PagerDuty

Created April 13, 2022 at 5:24 PM
Updated August 8, 2023 at 2:28 PM
Resolved August 3, 2022 at 8:49 PM