Define a problem statement

[Shreyanand]

As a data scientist, I would like to apply the time series introduction material to an actual problem statement. Comments and ideas are welcome.

• [ ] Acceptance: one valid problem statement to work on.

[MichaelClifford]

What is the purpose of this project? To connect with real Prometheus data, then demo a number of time series analysis techniques similar to the categorical encoding work? If so, wouldn't a single "problem statement" artificially narrow the scope? Have you defined what techniques/things you want to cover in the "time series introduction material" yet?

You might want to define a couple of problem statements like:

• Revisit some of the Prometheus anomaly detection work to explore a number of time-series forecasting methods.
• Look at this summit demo about encoding time-series signals into fixed length feature vectors for signal classification.
• Some basic signal modelling with linear and polynomial regressions.
• Or Spike/Anomaly Detection in time signals using 1d convolutions like was done in some of the CCX work .

I dunno. Just some ideas :smiley: WDYT?

[Shreyanand]

What is the purpose of this project? To connect with real Prometheus data, then demo a number of time series analysis techniques similar to the categorical encoding work? If so, wouldn't a single "problem statement" artificially narrow the scope? Have you defined what techniques/things you want to cover in the "time series introduction material" yet?

That is correct: to demo a number of time series analysis techniques and Prometheus data. However, with a single problem what I aimed was to show that we could use all these techniques in an example real project. I added this issue so that while I'm working on the demo notebooks, I could start looking at an example problem that I could use likely at the end of the demos but still as a part of the project. If that seems out of the scope, let me know, I'll re-frame this issue to track techniques.

You might want to define a couple of problem statements like:

• Revisit some of the Prometheus anomaly detection work to explore a number of time-series forecasting methods.
• Look at this summit demo about encoding time-series signals into fixed length feature vectors for signal classification.
• Some basic signal modelling with linear and polynomial regressions.
• Or Spike/Anomaly Detection in time signals using 1d convolutions like was done in some of the CCX work .

I dunno. Just some ideas WDYT?

These are great, I didn't know about the spike detection work and the summit demo, I'll look into it! Thanks for sharing :)

[MichaelClifford]

@Shreyanand can we close this issue for now?