Python Polars

Polars is a lightning-fast DataFrame library for Python built on Apache Arrow. It provides an expression-based API, lazy evaluation framework, and automatic parallelization for high-performance data processing.

Quick start

Installation

uv pip install polars

Basic operations

import polars as pl

# Create DataFrame
df = pl.DataFrame({
    "name": ["Alice", "Bob", "Charlie"],
    "age": [25, 30, 35],
    "city": ["NY", "LA", "SF"]
})

# Select columns
df.select("name", "age")

# Filter rows
df.filter(pl.col("age") > 25)

# Add computed columns
df.with_columns(
    age_plus_10=pl.col("age") + 10
)

Core concepts

Expressions

Expressions are composable units describing data transformations. Use pl.col("column_name") to reference columns and chain methods for complex operations:

df.select(
    pl.col("name"),
    (pl.col("age") * 12).alias("age_in_months")
)

Expressions execute within contexts: select(), with_columns(), filter(), group_by().agg().

Lazy vs eager evaluation

Eager (DataFrame): Operations execute immediately.

df = pl.read_csv("file.csv")  # Reads immediately
result = df.filter(pl.col("age") > 25)

Lazy (LazyFrame): Operations build an optimized query plan.

lf = pl.scan_csv("file.csv")  # Doesn't read yet
result = lf.filter(pl.col("age") > 25).select("name", "age")
df = result.collect()  # Executes optimized query

Use lazy mode for large datasets, complex pipelines, and when performance is critical. Benefits include automatic query optimization, predicate pushdown, projection pushdown, and parallel execution.

For detailed concepts including data types, type casting, null handling, and parallelization, see references/core-concepts.md.

Common operations

Select and with_columns

# Select specific columns
df.select("name", "age")

# Select with expressions
df.select(
    pl.col("name"),
    (pl.col("age") * 2).alias("double_age")
)

# Add new columns (preserves existing)
df.with_columns(
    age_doubled=pl.col("age") * 2,
    name_upper=pl.col("name").str.to_uppercase()
)

Filter

# Single condition
df.filter(pl.col("age") > 25)

# Multiple conditions (AND)
df.filter(
    pl.col("age") > 25,
    pl.col("city") == "NY"
)

# OR conditions
df.filter(
    (pl.col("age") > 25) | (pl.col("city") == "LA")
)

Group by and aggregations

df.group_by("city").agg(
    pl.col("age").mean().alias("avg_age"),
    pl.len().alias("count")
)

Window functions

Apply aggregations while preserving row count:

df.with_columns(
    avg_age_by_city=pl.col("age").mean().over("city"),
    rank_in_city=pl.col("salary").rank().over("city")
)

For comprehensive operations including sorting, conditionals, string/date operations, and list handling, see references/operations.md.

Data I/O

CSV

# Eager
df = pl.read_csv("file.csv")
df.write_csv("output.csv")

# Lazy (preferred for large files)
lf = pl.scan_csv("file.csv")
result = lf.filter(...).select(...).collect()

Parquet (recommended for performance)

df = pl.read_parquet("file.parquet")
df.write_parquet("output.parquet")

# Lazy with predicate pushdown
lf = pl.scan_parquet("file.parquet")

For comprehensive I/O including JSON, Excel, databases, cloud storage, and streaming, see references/io-guide.md.

Transformations

Joins

# Inner join
df1.join(df2, on="id", how="inner")

# Left join
df1.join(df2, on="id", how="left")

# Different column names
df1.join(df2, left_on="user_id", right_on="id")

Concatenation

# Vertical (stack rows)
pl.concat([df1, df2], how="vertical")

# Horizontal (add columns)
pl.concat([df1, df2], how="horizontal")

Pivot and unpivot

# Pivot (wide format)
df.pivot(values="sales", index="date", columns="product")

# Unpivot (long format)
df.unpivot(index="id", on=["col1", "col2"])

For detailed transformation patterns including asof joins, exploding, and transposing, see references/transformations.md.

Best practices

Performance optimization

1. Use lazy evaluation for large datasets:

   pythonCopy

   lf = pl.scan_csv("large.csv")  # Not read_csv
   result = lf.filter(...).select(...).collect()
   

2. Avoid Python functions in hot paths - stay within the expression API for parallelization:

   pythonCopy

   # Good: Native expression (parallelized)
   df.with_columns(result=pl.col("value") * 2)
   
   # Avoid: Python function (sequential)
   df.with_columns(result=pl.col("value").map_elements(lambda x: x * 2))
   

3. Select only needed columns early:

   pythonCopy

   lf.select("col1", "col2").filter(...)  # Good
   lf.filter(...).select("col1", "col2")  # Less optimal
   

4. Use streaming for very large data:

   pythonCopy

   lf.collect(streaming=True)
   

5. Use appropriate data types - Categorical for low-cardinality strings, appropriate integer sizes.

Conditional operations

pl.when(condition).then(value).otherwise(other_value)

Null handling

pl.col("x").fill_null(0)
pl.col("x").is_null()
pl.col("x").drop_nulls()

For comprehensive best practices including anti-patterns, memory management, testing, and code organization, see references/best-practices.md.

Pandas migration

Polars offers significant performance improvements over pandas with a cleaner API. Key differences:

• No index: Polars uses integer positions only
• Strict typing: No silent type conversions
• Lazy evaluation: Available via LazyFrame
• Parallel by default: Operations parallelized automatically

Common operation mappings

For comprehensive migration guide including operation mappings, migration patterns, and anti-patterns to avoid, see references/pandas-migration.md.

References

This skill includes comprehensive reference documentation:

• references/core-concepts.md - Expressions, data types, lazy evaluation, parallelization
• references/operations.md - Selection, filtering, grouping, window functions, string/date operations
• references/best-practices.md - Performance optimization, anti-patterns, memory management
• references/io-guide.md - CSV, Parquet, JSON, Excel, databases, cloud storage
• references/transformations.md - Joins, concatenation, pivots, reshaping operations
• references/pandas-migration.md - Migration guide from pandas to Polars

Load these references as needed for detailed information on specific topics.