Expressions beat map_elements in Python

map_elements drops into Python per row: correct, but the slow path.

import polars as pl

frame = pl.DataFrame({"tpm": [104, 88, 96]})

slow = frame.with_columns(
    pl.col("tpm").map_elements(lambda n: n * 2, return_dtype=pl.Int64)
      .alias("doubled_python"))

fast = frame.with_columns((pl.col("tpm") * 2).alias("doubled_expr"))

print(slow)
print(fast)
print(slow["doubled_python"].to_list() == fast["doubled_expr"].to_list())

How it works

1. An expression stays in Rust and parallelises.
2. map_elements needs a return dtype to stay typed.
3. Reach for it only when no expression exists.

Keywords and builtins used here

• as
• lambda
• print

The run, in numbers

Lines

13

Characters to type

355

Tokens

120

Three-star pace

115 tpm

At the three-star pace of 115 tokens a minute, this run takes about 63 seconds.

Type this snippet

Step 4 of 4 in Lazy & SQL, step 28 of 32 in Polars.

← Previous Next →