Example SampleStream Timings

The parameters you use when creating your samplestream depend on your application and what you're hunting for.

Are you looking for events that only show up at 20-millisecond resolution? You can scan long time frames at high resolution using settings like these…

These settings will scan the span range, delivering a snapshot of every 2 seconds with 100 samples (20-millisecond resolution).

In some cases you need to compensate for lag or find specific events, but the timing can be very erratic - the even might be seconds ago, or last week.

By using coarser resolutions, you can scan large periods of time for events. Generally, we'd suggest using a number of samples that is half the length of the event you want to capture - so if you're searching for an event that lasts 30 seconds, you should have at least 15-second resolution.

These settings will scan the span range, delivering a snapshot of every minute and data points every 15 seconds.

When training AI, it's often good to have a number of samples across a window of time, and for some of those samples to overlap.

By setting the step to smaller than the window, you can ensure that the next sample still contains some data from the current sample.

These settings will scan the span range, delivering 30 seconds of data at one-second resolution. Each sample will only move forward 15 seconds, meaning there will be 50% overlap from sample-to-sample.