# Duplicate Predictions

Some competitions have a duplicate detection feature that flags all models that have a prediction correlation above 95%[^1] and makes them ineligible for prizes.

The correlation is computed using [`pandas.DataFrame.corr(method="spearman")`](https://pandas.pydata.org/docs/reference/api/pandas.DataFrame.corr.html).

### Grouping

The prediction are first grouped between:

* user, to avoid duplication between multiple models
* team, to avoid duplicate between all models of every team member

The correlation function is then called for each prediction against each other.

### Keeping

Predictions are then grouped into correlated pairs to isolate them.

The best[^2] model in a pair is always kept. Other models are treated as copies.

#### Example

* If <mark style="color:blue;">A</mark> & <mark style="color:purple;">B</mark> are considered duplicates, and <mark style="color:orange;">C</mark> & <mark style="color:yellow;">D</mark> are also considered duplicates, then <mark style="color:blue;">A</mark> and <mark style="color:orange;">C</mark> will be retained and <mark style="color:purple;">B</mark> and <mark style="color:yellow;">D</mark> will have the duplicate badge.
* However, if <mark style="color:purple;">B</mark> & <mark style="color:orange;">C</mark> are also considered duplicates, then only <mark style="color:blue;">A</mark> will be retained. <mark style="color:purple;">B</mark>, <mark style="color:orange;">C</mark>, <mark style="color:yellow;">D</mark> will have the duplicate badge.
* In some cases, even if <mark style="color:blue;">A</mark> and <mark style="color:purple;">D</mark> are not considered to be duplicates, because of the link from <mark style="color:blue;">A</mark> to <mark style="color:purple;">B</mark> to <mark style="color:orange;">C</mark> to <mark style="color:yellow;">D</mark>, the duplicate badge is still displayed.

[^1]: This threshold can be changed at any time during the competition and at the discretion of the organizers.

[^2]: Old competition used the first model created.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.crunchdao.com/competitions/leaderboard/duplicate-predictions.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.