The simulation streams synthetic data points through a live t-digest, visualizing centroids forming and merging with denser clustering near the tails, then plots estimated percentiles against the true exact percentiles computed from the full dataset.
Choose a distribution shape and stream size, adjust the compression slider to change the centroid budget, and watch the error between estimated and exact percentiles shrink or grow as centroids are added or removed.
Distribution selector, stream size input, and a compression slider controlling the centroid budget.
A t-digest with only a couple hundred centroids can estimate the 99.9th percentile of a stream of a billion values with error small enough for production monitoring, using a memory footprint thousands of times smaller than storing the raw data.
The simulation streams synthetic data points through a live t-digest, visualizing centroids forming and merging with denser clustering near the tails, then plots estimated percentiles against the true exact percentiles computed from the full dataset.
The simulation streams synthetic data points through a live t-digest, visualizing centroids forming and merging with denser clustering near the tails, then plots estimated percentiles against the true exact percentiles computed from the full dataset.
Choose a distribution shape and stream size, adjust the compression slider to change the centroid budget, and watch the error between estimated and exact percentiles shrink or grow as centroids are added or removed.
A t-digest with only a couple hundred centroids can estimate the 99.9th percentile of a stream of a billion values with error small enough for production monitoring, using a memory footprint thousands of times smaller than storing the raw data.