[Python] Horizontal cum sum + unnest bug in polars

When I use horizontal cum sum followed by unnest, a "literal" column is formed that stays in the schema even when dropped.

Here is a mwe:

import polars as pl

def test_literal_bug():
print("Polars version:", pl.__version__)

# Create simple test data
df = pl.DataFrame({
"A": [1, 2, 3],
"T0": [0.1, 0.2, 0.3],
"T1": [0.4, 0.5, 0.6],
"T2": [0.7, 0.8, 0.9],
})

time_cols = ["T0", "T1", "T2"]
print("Original columns:", df.columns)
print("Time columns:", time_cols)

lazy_df = df.lazy()

print("Schema before cumsum:", lazy_df.collect_schema().names())

result = (
lazy_df.select(pl.cum_sum_horizontal(time_cols))
.unnest("cum_sum")
.rename({col: f"C{col}" for col in time_cols})
)

print("Schema after cumsum:", result.collect_schema().names())

# This will fail with: ColumnNotFoundError: "literal" not found
try:
collected = result.collect()
print("v1: No bug reproduced")
except pl.exceptions.ColumnNotFoundError as e:
print(f"v1: BUG REPRODUCED: {e}")

result_2 = result.drop("literal")
result_2 = pl.concat([pl.LazyFrame({"B": [1, 2, 3]}), result_2], how="horizontal")

print("Schema after drop and concat:", result_2.collect_schema().names())

try:
collected_2 = result_2.collect()
print("v2: No bug reproduced")
except pl.exceptions.ColumnNotFoundError as e:
print(f"v2: BUG REPRODUCED: {e}")

if __name__ == "__main__":
test_literal_bug()


which has output

Polars version: 1.31.0
Original columns: ['A', 'T0', 'T1', 'T2']
Time columns: ['T0', 'T1', 'T2']
Schema before cumsum: ['A', 'T0', 'T1', 'T2']
Schema after cumsum: ['CT0', 'CT1', 'CT2', 'literal']
v1: BUG REPRODUCED: "literal" not found
Schema after drop and concat: ['B', 'CT0', 'CT1', 'CT2']
v2: BUG REPRODUCED: "literal" not found


What is going on? Am I doing something wrong or is it a bug?

What follows is just to avoid the stackoverflow error "Too much code in your post": aaa

Continue reading...