ngram-polars 0.1.2

High-performance n-gram generation for Polars

ngram_polars - N-Gram Generation for Polars

A high-performance Polars plugin for generating n-grams from text data in Python.

Installation

pip install ngram-polars

Basic example

import polars as pl
from ngram_polars import ngrams

df = pl.DataFrame({
    "id": [1, 2],
    "words": [
        ["the", "quick", "brown", "fox"],
        ["hello", "world"]
    ]
})

# Generate bigrams
result = df.with_columns(
    bigrams=ngrams(pl.col("words"), n_range=[2])
)

more advanced examples

# Multiple n-gram sizes
df.with_columns(
    multi_ngrams=ngrams(pl.col("words"), n_range=[1, 2, 3])
)

# Custom delimiter
df.with_columns(
    underscored=ngrams(pl.col("words"), n_range=[2], delimiter="_")
)

# Lazy evaluation
(df.lazy()
   .with_columns(
       ngrams=ngrams(pl.col("words"), n_range=[2, 3])
   )
   .collect()
)

API Reference

ngrams(expr, n_range, delimiter) Generate n-grams from a list of strings.

Parameters:

• expr: IntoExpr - Polars expression representing a list of strings
• n_range: list[int] - List of n-gram sizes to generate (default: [1])
• delimiter: str - String delimiter between words (default: " ")

Returns:

• pl.Expr - Expression that generates lists of n-gram strings

Behavior:

• Returns a new list column containing all generated n-grams
• Works element-wise on list columns
• Changes the length of the output (each input list produces a new list of n-grams)
• Supports both eager and lazy evaluation

Performance Tips

• Use Lazy Evaluation: For large datasets, use lazy evaluation to optimize query planning
• Batch N-Gram Sizes: Generate multiple n-gram sizes in one call when possible
• Choose Appropriate N-Range: Only generate the n-gram sizes you actually need

Requirements

• Python 3.10 -> 3.14
• Polars requirement are not fully tested, tested on the latest version.
• Compatible with both eager and lazy Polars APIs

Changes

0.1.2: Updated to Rust 1.93.1 and Polars 0.53.0 (dropped python 3.9 added 3.14)