Trying to reduce the run time for a single time series

[gonzo-opera]

I'm using sigllm to do anomaly detection for a 3 hour timeseries with 1 minute interval between each data point (overall ~180 data points).

I'm using the openAI path using an API key. It take around couple of minutes to run for that time-series. Is there any way to make this faster? Is there any data on the fastest performance of sigllm?

Thanks!

[sarahmish]

Hi @gonzo-opera, thank you for opening an issue!

You should play around with the window_size and step_size of the pipeline. If you are working with 180 data points, I believe you don't need to even partition your data, you can set the window_size to fit all the data points.

You can also reduce the number of samples drawn by the LLM, currently the default is 10. However, having a larger number of samples will produce a more reliable result.

Let me know if you have any further questions!