Arrays Holding Multidimensional Windows of Streaming Data

[sirinath]

In dealing with vectorised time series it might be great to be able to efficiently represent such structures with both push and / or pull evaluation.

[SylvainCorlay]

@sirinath thanks for your feedback.

Can you elaborate on what you mean by push and pull evaluation?

[sirinath]

Push - when new data become available this triggers evaluation
Pull - when new items is needed (pulled) the rest is evaluated

Also ideally you should be able to have time based dimensions across where data is streamed.

[JohanMabille]

I think this should be implemented in a side-project as an extension to xtensor. A possible solution would be adapters on these times series that permits to plug them in the xtensor expression system and trigger the evaluation when needed.

@sirinath I close this issue since it's not something we plan to implement here, but I keep the idea in mind in case I have some time to implement it. In the meanwhile, feel free to propose an implementation if you have some time to do it.

[sirinath]

Is it possible to provide some rudimentary support for holding and computations on time series, vectorised time series data. I think this can be best implemented as array which has circular indexing. As new data comes in you append the vector along a dimension. This will be equivalent to numpy roll by 1 followed by updating the starting value but hopefully be more efficient for real time data. Also another equivalent would be to append and delete the oldest. I believe having a circular view along the axis would make both roll and this faster.