> For the complete documentation index, see [llms.txt](https://docs.earthindex.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.earthindex.ai/improve-your-results/quick-vs-deep-search.md).

# Quick Search or Deep Search

How Quick Search and Deep Search differ, when to use each, and how to combine them for better results.

Earth Index has two ways to search a project. Quick Search explores. Deep Search sweeps the whole area. This page helps you choose, and shows how to use them together.

## The short version

* Quick Search finds the tiles nearest to your examples. It does not train, so it is fast.
* Deep Search trains a classifier on your labels, then scores every tile in the area.
* Use Quick Search to improve your labels, then Deep Search for the final, complete list.

![Left: Quick Search circles the tiles nearest to your examples. Right: Deep Search learns a boundary and scores every tile in the area.](https://182773137-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FIQB2eToY1Iv2IH6cERjn%2Fuploads%2Fgit-blob-d9faa6445d9417473bcc77f593721667a98c45cd%2Fquick-vs-deep.svg?alt=media)

## How they differ

|                         | Quick Search                             | Deep Search                                                |
| ----------------------- | ---------------------------------------- | ---------------------------------------------------------- |
| What it does            | Finds the tiles nearest to your examples | Trains a classifier on your labels, then scores every tile |
| Learns from your labels | No training                              | Yes                                                        |
| Where it looks          | Inside the project area                  | Every tile in the project area                             |
| Results                 | 500 by default, up to 1,000              | Up to 100,000                                              |
| Speed                   | Fast                                     | Takes several minutes                                      |
| Minimum labels          | 1 positive                               | 10 positives and 1 negative                                |
| Access                  | Open and Full Access                     | Full Access                                                |

Quotas for each access level are in [Limits and quotas](https://docs.earthindex.ai/reference/limits-and-quotas.md).

## When to use each

**Use Quick Search to explore and to fix your labels.** It answers in seconds, so you can try a label, look at the top results and change your mind. It works with a single positive.

**Use Deep Search for the final sweep.** It uses all of your labels together and scores every tile, so it gives a ranked, complete list. It needs more labels and takes longer, so run it when your labels are in good shape.

## A working pattern

1. Label a few examples and run Quick Search.
2. Check the top results. Mark the wrong ones as negative and add missed examples as positive.
3. Run Quick Search again. Repeat until the top results are right.
4. Run Deep Search.
5. Review the top results of the Deep Search the same way. If a group of wrong results shows up, add negatives for it and run again.

Each Deep Search counts against your monthly allowance, so settle your labels with Quick Search first.

{% hint style="success" %}
**What to do**

Do not start with Deep Search. Iterate with Quick Search until the top results are right, then run Deep Search once.
{% endhint %}

<details>

<summary>Go deeper</summary>

Quick Search averages the embeddings of your positive labels into one query point. With negative labels, it moves that point away from their average. It then returns the tiles nearest to the point by cosine distance. Deep Search fits a logistic regression to the embeddings of your labelled tiles and scores every tile in the area. A tile is a detection when its probability is above the **Minimum Confidence** you set.

* [Logistic regression (scikit-learn)](https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression)

</details>

## Related

* [Why negatives matter](https://docs.earthindex.ai/improve-your-results/why-negatives-matter.md)
* [Deep Search](https://docs.earthindex.ai/search/deep-search.md)
* [Quick Search](https://docs.earthindex.ai/search/quick-search.md)
