examples

arnio Examples

Ready-to-run cleaning recipes for common messy CSV formats.

Each folder contains a small synthetic CSV with realistic data quality issues and a recipe.py that cleans, profiles, and validates it using the arnio API.

Recipes

Dataset	CSV	Issues Covered	arnio Features Used
Sales	messy_sales.csv	duplicates, mixed-case regions, missing revenue	pipeline, profile, validate
Customers	messy_customers.csv	whitespace in names, missing city, duplicate records	suggest_cleaning, pipeline, validate
Survey	messy_survey.csv	mixed-case labels, missing scores, duplicate submissions	profile, pipeline, validate
Logs	messy_logs.csv	mixed log levels, duplicate lines, missing timestamp	scan_csv, pipeline, profile, validate
Finance	messy_finance.csv	mixed-case types, duplicate transactions, missing amount	auto_clean, pipeline, validate

Running an Example

pip install arnio

cd examples/sales
python recipe.py

arnio API Features Demonstrated

• ar.read_csv() — load a CSV file into an ArFrame via the C++ backend
• ar.scan_csv() — inspect inferred column types without loading all data
• ar.profile() — generate a DataQualityReport (nulls, duplicates, whitespace, unique counts)
• ar.suggest_cleaning() — get auto-suggested pipeline steps from a profile
• ar.auto_clean() — one-call safe or strict cleaning with optional report return
• ar.pipeline() — apply a declarative sequence of cleaning steps
• ar.validate() — check an ArFrame against a typed Schema
• ar.to_pandas() — convert a cleaned ArFrame to a pandas DataFrame

Pipeline Steps Used

Step	Purpose
strip_whitespace	Trim leading/trailing whitespace from string columns
normalize_case	Standardise string values to lower, upper, or title case
drop_duplicates	Remove exact duplicate rows
fill_nulls	Replace nulls with a fallback value in specified columns
drop_nulls	Drop rows where specified columns are null

Schema Field Types

ar.String(nullable=False, allowed={"north", "south"})
ar.Int64(nullable=True, min=0, max=120)
ar.Float64(nullable=False, min=0.0)
ar.Email(nullable=False, unique=True)