# What is Rainbow? Definition and meaning

Combines prioritized replay, multi-step returns, distributional values, noisy networks, double and dueling learning in a single agent.

## Definition

**Rainbow**: Combines prioritized replay, multi-step returns, distributional values, noisy networks, double and dueling learning in a single agent.

In short: A combination of DQN improvements in one agent.

## When it is used

A strong value-based baseline for discrete-action reinforcement learning.

## Learn more

The full guide to **[Rainbow](/tools/rainbow)** explains how it works, when to use it and how it compares with the alternatives, with examples. The [glossary](/glossary) lists every term on the site from A to Z.

Previous term: [RAGAS](/glossary/ragas) · Next term: [Random forest](/glossary/random-forest)
