Skip to content

Reference

polars provides a large number of functions for numerous data types and this can sometimes be a bit overwhelming. Overall, you should be able to do anything you want with polars by specifying the data structure you want to use and then by applying expressions in a particular context.

Data structure

As explained in some vignettes, one of polars biggest strengths is the ability to choose between eager and lazy evaluation, that require respectively a DataFrame and a LazyFrame (with their counterparts GroupBy and LazyGroupBy for grouped data).

We can apply functions directly on a DataFrame or LazyFrame, such as rename() or drop(). Most functions that can be applied to DataFrames can also be used on LazyFrames, but some are specific to one or the other. For example:

  • $equals() exists for DataFrame but not for LazyFrame;
  • $collect() executes a lazy query, which means it can only be applied on a LazyFrame.

Another common data structure is the Series, which can be considered as the equivalent of R vectors in polars’ world. Therefore, a DataFrame is a list of Series.

Operations on DataFrame or LazyFrame are useful, but many more operations can be applied on columns themselves by using various expressions in different contexts.

Contexts

A context simply is the type of data modification that is done. There are 3 types of contexts:

  • select and modify columns with select() and with_columns();
  • filter rows with filter();
  • group and aggregate rows with group_by() and agg()

Inside each context, you can use various expressions (aka. Expr). Some expressions cannot be used in some contexts. For example, in with_columns(), you can only apply expressions that return either the same number of values or a single value that will be duplicated on all rows:

test <- as_polars_df(mtcars)

# this works
test$with_columns(pl$col("mpg") + 1)
#> shape: (32, 11)
#> ┌──────┬─────┬───────┬───────┬───┬─────┬─────┬──────┬──────┐
#> │ mpg  ┆ cyl ┆ disp  ┆ hp    ┆ … ┆ vs  ┆ am  ┆ gear ┆ carb │
#> │ ---  ┆ --- ┆ ---   ┆ ---   ┆   ┆ --- ┆ --- ┆ ---  ┆ ---  │
#> │ f64  ┆ f64 ┆ f64   ┆ f64   ┆   ┆ f64 ┆ f64 ┆ f64  ┆ f64  │
#> ╞══════╪═════╪═══════╪═══════╪═══╪═════╪═════╪══════╪══════╡
#> │ 22.0 ┆ 6.0 ┆ 160.0 ┆ 110.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 4.0  ┆ 4.0  │
#> │ 22.0 ┆ 6.0 ┆ 160.0 ┆ 110.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 4.0  ┆ 4.0  │
#> │ 23.8 ┆ 4.0 ┆ 108.0 ┆ 93.0  ┆ … ┆ 1.0 ┆ 1.0 ┆ 4.0  ┆ 1.0  │
#> │ 22.4 ┆ 6.0 ┆ 258.0 ┆ 110.0 ┆ … ┆ 1.0 ┆ 0.0 ┆ 3.0  ┆ 1.0  │
#> │ 19.7 ┆ 8.0 ┆ 360.0 ┆ 175.0 ┆ … ┆ 0.0 ┆ 0.0 ┆ 3.0  ┆ 2.0  │
#> │ …    ┆ …   ┆ …     ┆ …     ┆ … ┆ …   ┆ …   ┆ …    ┆ …    │
#> │ 31.4 ┆ 4.0 ┆ 95.1  ┆ 113.0 ┆ … ┆ 1.0 ┆ 1.0 ┆ 5.0  ┆ 2.0  │
#> │ 16.8 ┆ 8.0 ┆ 351.0 ┆ 264.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 4.0  │
#> │ 20.7 ┆ 6.0 ┆ 145.0 ┆ 175.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 6.0  │
#> │ 16.0 ┆ 8.0 ┆ 301.0 ┆ 335.0 ┆ … ┆ 0.0 ┆ 1.0 ┆ 5.0  ┆ 8.0  │
#> │ 22.4 ┆ 4.0 ┆ 121.0 ┆ 109.0 ┆ … ┆ 1.0 ┆ 1.0 ┆ 4.0  ┆ 2.0  │
#> └──────┴─────┴───────┴───────┴───┴─────┴─────┴──────┴──────┘

# this doesn't work because it returns only 2 values, while mtcars has 32 rows.
test$with_columns(pl$col("mpg")$slice(0, 2))

By contrast, in an agg() context, any number of return values are possible, as they are returned in a list, and only the new columns or the grouping columns are returned.

test$group_by(pl$col("cyl"))$agg(
  pl$col("mpg"), # varying number of values
  pl$col("mpg")$slice(0, 2)$name$suffix("_sliced"), # two values
  # aggregated to one value and implicitly unpacks list
  pl$col("mpg")$sum()$name$suffix("_summed")
)
#> shape: (3, 4)
#> ┌─────┬──────────────────────┬──────────────┬────────────┐
#> │ cyl ┆ mpg                  ┆ mpg_sliced   ┆ mpg_summed │
#> │ --- ┆ ---                  ┆ ---          ┆ ---        │
#> │ f64 ┆ list[f64]            ┆ list[f64]    ┆ f64        │
#> ╞═════╪══════════════════════╪══════════════╪════════════╡
#> │ 6.0 ┆ [21.0, 21.0, … 19.7] ┆ [21.0, 21.0] ┆ 138.2      │
#> │ 8.0 ┆ [18.7, 14.3, … 15.0] ┆ [18.7, 14.3] ┆ 211.4      │
#> │ 4.0 ┆ [22.8, 24.4, … 21.4] ┆ [22.8, 24.4] ┆ 293.3      │
#> └─────┴──────────────────────┴──────────────┴────────────┘

Expressions

Expressions are the building blocks that give all the flexibility we need to modify or create new columns.

Two important expressions starters are pl$col() (names a column in the context) and pl$lit() (wraps a literal value or vector/series in an Expr). Most other expression starters are syntactic sugar derived from thereof, e.g. pl$sum(_) is actually pl$col(_)$sum().

Expressions can be chained with more than 170 expression methods such as $sum() which aggregates e.g. the column with summing.

# two examples of starting, chaining and combining expressions
pl$DataFrame(a = 1:4)$with_columns(
  # compute the cosine of column "a"
  a_cos = pl$col("a")$cos()$sin(),
  # standardize the values of column "a"
  a_stand = (pl$col("a") - pl$col("a")$mean()) / pl$col("a")$std(),
  # take 1:3, name it, then sum, then multiply by two
  lit_sum_add_two = pl$lit(1:3)$sum() * 2L
)
#> shape: (4, 4)
#> ┌─────┬───────────┬───────────┬─────────────────┐
#> │ a   ┆ a_cos     ┆ a_stand   ┆ lit_sum_add_two │
#> │ --- ┆ ---       ┆ ---       ┆ ---             │
#> │ i32 ┆ f64       ┆ f64       ┆ i32             │
#> ╞═════╪═══════════╪═══════════╪═════════════════╡
#> │ 1   ┆ 0.514395  ┆ -1.161895 ┆ 12              │
#> │ 2   ┆ -0.404239 ┆ -0.387298 ┆ 12              │
#> │ 3   ┆ -0.836022 ┆ 0.387298  ┆ 12              │
#> │ 4   ┆ -0.608083 ┆ 1.161895  ┆ 12              │
#> └─────┴───────────┴───────────┴─────────────────┘

Some methods share a common name but their behavior might be very different depending on the input type. For example, $decode() doesn’t do the same thing when it is applied on binary data or on string data.

To be able to distinguish those usages and to check the validity of a query, polars stores methods in subnamespaces. For each datatype other than numeric (floats and integers), there is a subnamespace containing the available methods: dt (datetime), list (list), str (strings), struct (structs), cat (categoricals) and bin (binary). As a sidenote, there is also an exotic subnamespace called meta which is rarely used to manipulate the expressions themselves. Each subsection in the “Expressions” section lists all operations available for a specific subnamespace.

For a concrete example for dt, suppose we have a column containing dates and that we want to extract the year from these dates:

# Create the DataFrame
df <- pl$select(
  date = pl$date_range(
    as.Date("2020-01-01"),
    as.Date("2023-01-02"),
    interval = "1y"
  )
)
df
#> shape: (4, 1)
#> ┌────────────┐
#> │ date       │
#> │ ---        │
#> │ date       │
#> ╞════════════╡
#> │ 2020-01-01 │
#> │ 2021-01-01 │
#> │ 2022-01-01 │
#> │ 2023-01-01 │
#> └────────────┘

The function year() only makes sense for date-time data, so we look for functions in the dt subnamespace (for date-time):

df$with_columns(year = pl$col("date")$dt$year())
#> shape: (4, 2)
#> ┌────────────┬──────┐
#> │ date       ┆ year │
#> │ ---        ┆ ---  │
#> │ date       ┆ i32  │
#> ╞════════════╪══════╡
#> │ 2020-01-01 ┆ 2020 │
#> │ 2021-01-01 ┆ 2021 │
#> │ 2022-01-01 ┆ 2022 │
#> │ 2023-01-01 ┆ 2023 │
#> └────────────┴──────┘

Similarly, to convert a string column to uppercase, we use the str prefix before using to_uppercase():

# Create the DataFrame
pl$DataFrame(foo = c("jake", "mary", "john peter"))$with_columns(
  upper = pl$col("foo")$str$to_uppercase()
)
#> shape: (3, 2)
#> ┌────────────┬────────────┐
#> │ foo        ┆ upper      │
#> │ ---        ┆ ---        │
#> │ str        ┆ str        │
#> ╞════════════╪════════════╡
#> │ jake       ┆ JAKE       │
#> │ mary       ┆ MARY       │
#> │ john peter ┆ JOHN PETER │
#> └────────────┴────────────┘