pandasclean 0.1.2

A lightweight library for cleaning and optimizing pandas DataFrames

pandasclean

A lightweight Python library for cleaning and optimizing pandas DataFrames. Built for data analysts and data scientists who want practical, no-fuss data cleaning with sensible defaults.

---

Quick Start

import pandas as pd
from pandasclean import auto_clean

df = pd.read_csv('your_data.csv')
df_clean, report = auto_clean(df)

That's it. One line cleans your entire DataFrame.

---

Features

• Outlier Detection & Handling — Detect outliers using the IQR method and choose to report, drop, or cap them
• Memory Reduction — Automatically downcast numeric dtypes and convert low cardinality string columns to save memory
• NaN Handling — Drop, fill with mean/median, or supply custom fill values per column
• Auto Clean — One function that runs everything with sensible defaults

---

Installation

pip install pandasclean

---

Usage

Auto Clean

Runs all cleaning functions in the correct order with sensible defaults.

from pandasclean import auto_clean

df_clean, report = auto_clean(df)

For custom behaviour, use the individual functions directly.

---

Outlier Detection & Handling

from pandasclean import find_outliers

# Report outlier bounds without changing data
df, bounds = find_outliers(df, strategy='report')

# Drop rows containing outliers
df_clean, bounds = find_outliers(df, strategy='drop')

# Cap outliers to the nearest bound (Winsorization)
df_clean, bounds = find_outliers(df, strategy='cap')

# Target specific columns with a custom multiplier
df_clean, bounds = find_outliers(df, columns=['age', 'salary'], multiplier=3.0, strategy='cap')

---

NaN Handling

from pandasclean import handle_nan

# Report null counts and percentages without making changes
df, report = handle_nan(df, strategy='report')

# Drop rows with any NaN
df_clean, report = handle_nan(df, strategy='drop', axis='rows', how='any')

# Drop columns where more than 50% of values are NaN
df_clean, report = handle_nan(df, strategy='drop', axis='columns', threshold=50)

# Fill NaN with column mean (numeric columns only)
df_clean, report = handle_nan(df, strategy='mean')

# Fill NaN with column median (numeric columns only)
df_clean, report = handle_nan(df, strategy='median')

# Fill all NaN with a single custom value
df_clean, report = handle_nan(df, strategy='custom', fill_value=0)

# Fill NaN with different values per column
df_clean, report = handle_nan(df, strategy='custom', fill_value={
    'age': 0,
    'name': 'unknown',
    'salary': 50000
})

---

Memory Reduction

from pandasclean import reduce_memory

# Optimize all columns with default settings
df_optimized, report = reduce_memory(df)

# Disable category conversion for string columns
df_optimized, report = reduce_memory(df, convert_category=False)

# Custom cardinality threshold for category conversion
df_optimized, report = reduce_memory(df, cardinality_threshold=0.3)

Benchmarks

Tested on a 1.5 million row DataFrame with mixed dtypes:

Column	Before	After	Notes
Account_ID	int64	int32	Downcast
Customer_Age	int64	int8	Downcast
Account_Balance	float64	float32	Downcast
Monthly_Spend	float64	float32	Downcast
Credit_Score	float64	float32	Downcast
Activity_Metric_1	float64	float32	Downcast
Activity_Metric_2	float64	float32	Downcast
Tier	object	category	Low cardinality

before = df.memory_usage(deep=True).sum() / (1024 * 1024)
df_optimized, report = reduce_memory(df)
after = df_optimized.memory_usage(deep=True).sum() / (1024 * 1024)

# Before: 152.71 MB
# After:  37.19 MB
# Reduction: 75.6%

Always use memory_usage(deep=True) for accurate measurement — pandas default undercounts string column memory.

---

How It Works

Memory Reduction

dtype	Action
int64	Downcast to smallest safe type (int8 → int16 → int32)
float64	Downcast to float32 where possible
object / str	Convert to category if cardinality ratio is below threshold

Outlier Detection

Uses the IQR method to compute bounds:

• lower_bound = Q1 - (multiplier × IQR)
• upper_bound = Q3 + (multiplier × IQR)

Standard multiplier values:

• 1.5 — mild outliers (default)
• 3.0 — extreme outliers only

---

Parameters

auto_clean(df)

Parameter	Default	Description
df	required	Input DataFrame

---

find_outliers(df, columns, multiplier, strategy)

Parameter	Default	Description
df	required	Input DataFrame
columns	None	Columns to check. Defaults to all numeric columns
multiplier	1.5	IQR multiplier. Use 3.0 for extreme outliers only
strategy	'report'	One of 'report', 'drop', 'cap'

---

handle_nan(df, columns, strategy, fill_value, axis, how, threshold)

Parameter	Default	Description
df	required	Input DataFrame
columns	None	Columns to process. Defaults to all columns
strategy	'report'	One of 'report', 'drop', 'mean', 'median', 'custom'
fill_value	None	Scalar or dict. Required when strategy is 'custom'
axis	'rows'	'rows' or 'columns'. Only used with 'drop' strategy
how	'any'	'any' or 'all'. Only used with 'drop' strategy
threshold	None	NaN percentage threshold for dropping. Overrides how when set

---

reduce_memory(df, columns, convert_category, cardinality_threshold)

Parameter	Default	Description
df	required	Input DataFrame
columns	None	Columns to process. Defaults to all columns
convert_category	True	Whether to convert low cardinality strings to category
cardinality_threshold	0.5	Max unique ratio to trigger category conversion

---

Use Cases

• Data cleaning before analysis — handle outliers and NaN values in one step
• Memory optimization before ML training — reduce DataFrame size before fitting models, especially useful for GPU training where float32 is standard
• CSV/Parquet compression — smaller dtypes mean smaller files on disk. Saving to parquet after running reduce_memory can reduce file sizes significantly

---

Roadmap

• Outlier detection and handling (IQR method)
• Memory reduction (dtype downcasting + category conversion)
• NaN handling (drop, mean, median, custom)
• Auto clean
• Z-score based outlier detection
• Skewness detection and fixing
• Duplicate detection and removal
• HTML report generator

---

License

MIT License