# percentile

> For the complete documentation index, see [llms.txt](https://kotlin.github.io/dataframe/llms.txt).

Computes the specified [percentile](https://en.wikipedia.org/wiki/Percentile) of values.

API reference: [percentile](https://kotlin.github.io/dataframe/api-docs/core/org.jetbrains.kotlinx.dataframe.api/percentile.html)

This is also called the "centile" or the 100-[quantile](https://en.wikipedia.org/wiki/Quantile).

The 25th percentile is also known as the first [quartile](https://en.wikipedia.org/wiki/Quartile) (Q1), the 50th percentile as the [median](median.html) or second [quartile](https://en.wikipedia.org/wiki/Quartile) (Q2), and the 75th percentile as the third [quartile](https://en.wikipedia.org/wiki/Quartile) (Q3).

`null` values in the input are ignored. The operations either throw an exception when the input is empty (after filtering `null` or `NaN` values), or they return `null` when using the `-orNull` overloads.

All primitive numeric types are supported: `Byte`, `Short`, `Int`, `Long`, `Float`, and `Double`, but no mix of different number types. In these cases, the return type is always `Double?`. The results of the operation on these types are interpolated using [Quantile Estimation Method](#quantile-estimation-methods) R8.

The operation is also available for self-comparable columns (so columns of type `T : Comparable<T>`, whose values are mutually comparable, like `DateTime`, `String`, etc.) In this case, the return type remains `T?`. The index of the result of the operation on these types is rounded using [Quantile Estimation Method](#quantile-estimation-methods) R3.

All operations on `Double`/`Float` have the `skipNaN` option, which is set to `false` by default. This means that if a `NaN` is present in the input, it will be propagated to the result. When it's set to `true`, `NaN` values are ignored.

## Quantile Estimation Methods

DataFrame follows [Hyndman, Rob &amp; Fan, Yanan. (1996). Sample Quantiles in Statistical Packages. The American Statistician. 50. 361-365. 10.1080/00031305.1996.10473566.](https://www.researchgate.net/publication/222105754_Sample_Quantiles_in_Statistical_Packages) and [Apache Commons Statistics](https://svn.apache.org/repos/infra/websites/production/commons/content/proper/commons-statistics/commons-statistics-docs/apidocs/org/apache/commons/statistics/descriptive/Quantile.EstimationMethod.html) for the [9 commonly used quantile estimation methods](https://en.wikipedia.org/wiki/Quantile#Estimating_quantiles_from_a_sample).

For the `percentile` operation, DataFrame uses estimation method R3 when the given percentile needs to be selected from the values (like for self-comparable columns), and R8 when the given percentile can be interpolated from the values (of a numeric column). R8 was the recommended method by Hyndman and Fan, though other libraries, like [Numpy](https://numpy.org/doc/2.1/reference/generated/numpy.quantile.html) default to R7, so slightly different results are to be expected.

In the future we might add an option to change the quantile estimation method.

```KOTLIN
df.percentile(25.0) // 25th percentile of values for every column with mutually comparable values
df.percentile(75.0) { age and weight } // 75th percentile of all values in `age` and `weight`
df.percentileFor(50.0, skipNaN = true) { age and name.firstName } // 50th percentile of values per `age` and `firstName` separately
df.percentileOf(75.0) { (weight ?: 0) / age } // 75th percentile of expression evaluated for every row
df.percentileBy(25.0) { age } // DataRow where the 25th percentile of `age` lies (index rounded using R3)
```

```KOTLIN
df.percentile(25.0)
df.age.percentile(75.0)
df.groupBy { city }.percentile(50.0)
df.pivot { city }.percentile(75.0)
df.pivot { city }.groupBy { name.lastName }.percentile(25.0)
```

See [statistics](summarystatistics.html#groupby-statistics) for details on complex data aggregations.

See [column selectors](columnselectors.html) for how to select the columns for this operation.

## Type Conversion

The following automatic type conversions are performed for the `percentile` operation. (Note that `null` only appears in the return type when using `-orNull` overloads).

| Conversion | Result for Empty Input |
| --- | --- |
| T -> T where T : Comparable\<T\> | null |
| Int -> Double | null |
| Byte -> Double | null |
| Short -> Double | null |
| Long -> Double | null |
| Double -> Double | null |
| Float -> Double | null |
| Nothing -> Nothing | null |

## Big numbers

> **Warning:**
> `java.math.BigDecimal` and `java.math.BigInteger` are not explicitly supported. They are self-comparable, so some overloads (like `df.percentile(25.0)`) still select a value without interpolating it, while others (like `df.percentile(25.0) { bigDecimalCol }`) throw at runtime. Don't rely on this: compute the percentile manually with Kotlin standard library methods and Java big number arithmetic, or [convert](convert.html) the column to a primitive type first.

For a `BigDecimal` column `amount`, the exact percentile can be selected from the sorted values with the same [closest rank method](#quantile-estimation-methods) R3 that DataFrame uses:

```KOTLIN
// exact 25th percentile: the closest rank (R3) of the sorted values
val amounts = df.amount.toList().sorted()
val rank = round(0.25 * amounts.size).toInt().coerceIn(1, amounts.size)
amounts[rank - 1]
```

Alternatively, [convert](convert.html) the column to a primitive type first — this interpolates the percentile, at the cost of precision:

```KOTLIN
// approximate, interpolated 25th percentile, computed after converting the column to `Double`
df.convert { amount }.toDouble().percentile(25.0) { amount }
```