Introduction

Verdict is a lightweight system that can accurately answer your analytical queries. Instead of aggregating every value in your data, Verdict predicts the answers using its concise data summary, or equivalently samples. To maximize the efficiency, verdict incorporates advanced types of samples.

Verdict is designed to operate on top of existing analytics engines (e.g., Facebook Presto, Google BigQuery, Amazon Redshift). Thus, no data migration is required.

Verdict has two modes of operations:

  1. stream: Verdict gives you a series of answers as continuously processing data. Each answer is statistically unbiased. Internally, Verdict performs optimization to maximize the accuracy of the answers. Typically, even the first answer (available in less than a second) is almost identical to the exact one, however large your data is.
  2. traditional: Verdict gives you a single answer that satisfies a requested accuracy level (1% relative error by default). Again, Verdict’s latency doesn’t increase however large you data is.

Note

We are refactoring our old code; thus, some of the features mentioned in our website may not be available yet.