Fast data preparation for the Python data stack.

Arnio is a compiled C++ data preparation engine for messy CSV and pandas workflows.
It parses, infers types, strips whitespace, deduplicates, validates, and profiles data —
then hands clean results back to the tools you already use.
Use Arnio before and alongside pandas, NumPy, scikit-learn, DuckDB, and Arrow.

           

pip install arnio

Colab install smoke test: COLAB_SMOKE_TEST.md

Quickstart · Integrations · Why Arnio · Architecture · Benchmarks · Community · Contribute

---

⚡ Quickstart

A simple workflow in just a few steps.

New to Arnio? Start with the pandas workflow example below before exploring advanced pipelines.

import arnio as ar

# Load CSV directly through C++ — no Python parsing overhead
frame = ar.read_csv("messy_sales_data.csv")

# Strict mode (default) fails on inconsistent row widths
frame = ar.read_csv("messy_sales_data.csv", mode="strict")

# Permissive mode fills missing trailing values with nulls
frame = ar.read_csv("messy_sales_data.csv", mode="permissive")

# Declare what clean data looks like — arnio handles the rest
clean = ar.pipeline(frame, [
    ("strip_whitespace",),
    ("normalize_case", {"case_type": "lower"}),
    ("fill_nulls", {"value": 0.0, "subset": ["revenue"]}),
    ("drop_nulls",),
    ("drop_duplicates",),
])



# Out comes a standard pandas DataFrame — use it like you always have
df = ar.to_pandas(clean)

# Use copy=True when you need defensive pandas-owned buffers
safe_df = ar.to_pandas(clean, copy=True)

Dry Run Validation

Use dry_run=True to validate pipeline configuration and step execution without returning transformed output.

ar.pipeline(
    frame,
    [
        ("drop_nulls",),
    ],
    dry_run=True,
)

Need step timings for debugging? Opt in without changing the default pipeline return type:

clean, metadata = ar.pipeline(
    frame,
    [("strip_whitespace",), ("drop_duplicates",)],
    return_metadata=True,
)

print(metadata["step_timings"])
print(metadata["applied_steps"])
print(metadata["row_counts"])

Quick Example

import arnio

frame = arnio.read_csv("sample.csv")

# Preview first 5 rows
frame.preview(5)

Pipeline validation behavior

Pipeline step specifications are validated before execution begins.

Malformed step tuples, invalid kwargs structures, or unknown step names fail early before any pipeline steps execute.

ar.pipeline(
    frame,
    [
        ("strip_whitespace",),
        ("bad_step", "oops", "extra"),
    ],
)

This prevents partial pipeline execution when later pipeline steps are invalid.

Already have a pandas DataFrame? Use Arnio in-place in your existing pandas workflow:

import pandas as pd
import arnio as ar

df = pd.read_csv("messy_sales_data.csv")

clean_df = df.arnio.clean([
    ("strip_whitespace",),
    ("normalize_case", {"case_type": "lower"}),
    ("drop_duplicates",),
])

report = clean_df.arnio.profile()

Cross-field validation rules

Pass a rules list to Schema for checks that span multiple columns. Each rule receives the full pandas DataFrame and must return a list[ValidationIssue] — an empty list means the rule passed.

import arnio as ar

def end_after_start(df):
    return [
        ar.ValidationIssue(
            column="end_date",
            rule="cross_field",
            message="end_date must be >= start_date",
            row_index=int(i) + 1,
        )
        for i, row in df.iterrows()
        if row["end_date"] < row["start_date"]
    ]

schema = ar.Schema(
    {"start_date": ar.String(), "end_date": ar.String()},
    rules=[end_after_start],
)

result = schema.validate(ar.read_csv("events.csv"))
print(result.passed)

Row index convention: ValidationIssue.row_index values are 1-based and count data rows only. The header row is excluded. row_index=1 is the first data row in the file.

Schema diff reports

Use diff_schema() to compare expected and observed data contracts across datasets, releases, or generated schemas.

import arnio as ar

expected = ar.Schema({
    "id": ar.Int64(nullable=False, unique=True),
    "email": ar.Email(nullable=False),
})

observed = ar.Schema({
    "id": ar.Int64(nullable=False),
    "created_at": ar.DateTime(format="%Y-%m-%d"),
})

diff = ar.diff_schema(expected, observed)
print(diff.summary())
print(diff.to_markdown())

CI data contracts (GitHub Actions)

If you want to block schema drift or invalid rows in pull requests, see DATA_CONTRACT_CI.md for an inert copy-paste GitHub Actions workflow example.

Example contract files are included under examples/contracts/.

Select specific columns

Use select_columns() to create a new ArFrame with only the required columns before converting to pandas.

selected = frame.select_columns(["name", "revenue"])

print(selected.columns)
# ['name', 'revenue']

Handling missing values

Arnio supports configuring which strings are treated as null during CSV parsing using the null_values parameter in read_csv and scan_csv. By default, Arnio preserves its existing behavior and treats only empty cells as null. Custom matching is case-insensitive and applies to cell values only (not headers).

# Default behavior: empty cells are null
frame = ar.read_csv("data.csv")

# Provide a custom list of sentinels (overrides the empty-cell default)
frame = ar.read_csv("data.csv", null_values=["", "MISSING", "UNKNOWN"])

# Disable null sentinel handling completely
frame = ar.read_csv("data.csv", null_values=[])

Every step above executes in C++. Your Python code is a configuration — not the execution engine.

Explore more in the examples/ folder — ready-to-run recipes for sales, customers, survey, logs, and finance datasets.

Security note: CSV formula injection

Arnio preserves cell values when reading CSV files. It does not rewrite strings that begin with spreadsheet formula prefixes such as =, +, -, or @.

If you export Arnio-cleaned data back to CSV and expect users to open that file in Excel, Google Sheets, LibreOffice, or another spreadsheet application, treat untrusted text fields as potentially executable spreadsheet formulas. Before exporting, escape or neutralize formula-like strings in user-controlled columns, for example by prefixing a single quote or another project-approved escape marker.

This is especially important for customer names, notes, comments, imported form fields, and any other free-text values that may come from outside your trust boundary. Arnio focuses on parsing, validation, profiling, and cleanup; final CSV export policy should stay explicit in the application that writes the file.

Error Handling

read_csv and scan_csv

Input	Raises	Message
File not found	CsvReadError	Cannot open file: <path>
Zero-byte file	CsvReadError	CSV file is empty: '<path>'
Blank header line	CsvReadError	CSV header contains an empty column name
Binary / NUL bytes	CsvReadError	CSV input contains NUL bytes and appears to be binary or corrupted

Schema Validation

ar.validate() returns a ValidationResult; it does not raise for validation failures. Check result.passed and result.issues for dtype or required_column rule violations.

Pipeline Step Errors

Unknown step names raise UnknownStepError before execution begins.

📸 Peek at a 100 GB file without loading it

scan_csv reads only the header + a sample to infer the schema. Zero data loaded.

# Pass sample_size to control how many rows are evaluated for type inference
schema = ar.scan_csv("100GB_file.csv", sample_size=500)
# {'id': 'int64', 'name': 'string', 'is_active': 'bool', 'revenue': 'float64'}

Useful for exploring datasets before committing memory.

📄 Read JSON Lines (JSONL / NDJSON) files

read_jsonl parses one JSON object per line into an ArFrame. Blank lines are skipped, missing keys become nulls, and mixed-type columns are coerced to string — the same rules as from_pandas.

# events.jsonl
# {"user": "alice", "score": 9.5, "active": true}
# {"user": "bob",   "score": 8.1, "active": false}

frame = ar.read_jsonl("events.jsonl")

# Limit rows
frame = ar.read_jsonl("large.jsonl", nrows=1000)

# Non-UTF-8 encoding
frame = ar.read_jsonl("data.ndjson", encoding="latin-1")

# Plug straight into the cleaning pipeline
clean = ar.pipeline(frame, [("strip_whitespace",), ("drop_nulls",)])

Raises ar.JsonlReadError with the 1-based line number if a line contains invalid JSON.

📦 Export to Parquet for columnar analytics pipelines

write_parquet exports an ArFrame to a Parquet file via pyarrow. Install the optional extra first:

pip install arnio[parquet]

# Basic export
ar.write_parquet(frame, "output.parquet")

# Choose compression codec: "snappy" (default), "gzip", "zstd", "brotli", "none"
ar.write_parquet(frame, "output.parquet", compression="zstd")

# Control row group size for large files
ar.write_parquet(frame, "output.parquet", row_group_size=50_000)

# .pq extension also accepted
ar.write_parquet(frame, "output.pq")

Raises ImportError with an install hint if pyarrow is not available.

👀 Preview rows without pandas conversion or full-column Python list materialization

preview() reads only the first n rows directly from the C++ frame — no pandas conversion triggered.

frame = ar.read_csv("huge_file.csv")

print(frame.preview())      # first 5 rows (default)
print(frame.preview(n=10))  # first 10 rows

Raises ValueError for invalid n (zero, negative, or non-integer).

💰 Financial Decimal Support

arnio provides support for converting Python decimal.Decimal objects.

• Behavior: Python Decimal objects are automatically preserved as high-precision strings during serialization/binding to prevent floating-point precision loss.
• Caveat: When reading back into Pandas, to_pandas() returns these as string (object dtype) columns. You will need to explicitly cast them back to Decimal objects on the resulting DataFrame if you want to resume exact math.

Example:

from decimal import Decimal

import pandas as pd

import arnio as ar

df = pd.DataFrame({
    "price": [Decimal("19.99"), Decimal("29.95")]
})

frame = ar.from_pandas(df)  # Decimal values safely preserved as exact strings
result = ar.to_pandas(frame)
# result["price"] will be string objects ["19.99", "29.95"]

🧩 Add custom steps without touching C++

Register any Python function as a pipeline step. It receives a DataFrame, returns a DataFrame.

def remove_outliers(df, column="revenue", threshold=100_000):
    return df[df[column] <= threshold]

ar.register_step("remove_outliers", remove_outliers)
ar.register_step("team:drop_nulls", remove_outliers)  # namespaced custom step

# Use builtin: for an explicit built-in step, and your own prefixes
# like team: or plugin_name: to avoid name collisions.

# Introspect built-in and custom step names without reaching into internals.
print(ar.list_steps())

# Now use it in any pipeline alongside native C++ steps
clean = ar.pipeline(frame, [
    ("builtin:strip_whitespace",),
    ("remove_outliers", {"column": "revenue", "threshold": 50000}),
    ("drop_duplicates",),
])

Need to inspect the built-in kwargs a step accepts before assembling a pipeline?

signatures = ar.get_builtin_step_signatures()
print(list(signatures["drop_nulls"].parameters))  # ["subset"]
print(list(signatures["filter_rows"].parameters))  # ["column", "op", "value"]

Need to restore the registry back to built-in steps only during tests?

ar.reset_steps()

print(ar.list_steps())
# Only built-in steps remain

Custom steps run through a pandas↔ArFrame conversion bridge. Prototype in Python, then optionally migrate hot paths to C++ for full speed.

🔄 Custom Step Overwrite Policy

By default, trying to register a custom step with a name that is already taken by another custom Python step will raise a ValueError to prevent silent overwriting.

To intentionally replace an existing custom Python step, pass overwrite=True:

def custom_logging(df):
    print("Running step v1")
    return df

ar.register_step("log_data", custom_logging)

# This will succeed and safely overwrite the original logic
def custom_logging_v2(df):
    print("Running step v2")
    return df

ar.register_step("log_data", custom_logging_v2, overwrite=True)

Note: Built-in C++ pipeline steps (like "drop_nulls") can never be overwritten, even if overwrite=True is explicitly supplied.

✂️ Slice rows with head() and tail()

head() and tail() return the first or last n rows as a new ArFrame.

frame = ar.read_csv("data.csv")

frame.head()     # first 5 rows (default)
frame.head(10)   # first 10 rows
frame.tail(3)    # last 3 rows

# n larger than row count returns all rows safely
frame.head(1000)

# n=0 returns an empty ArFrame
frame.head(0)

Raises ValueError for negative or boolean n.

---

🔗 Integrations

Arnio is designed to make the rest of the Python data stack more productive, not to replace it.

Workflow	How Arnio helps
pandas	Clean, validate, and profile messy DataFrames through df.arnio.
NumPy	Prepare typed numeric data before array/modeling workflows.
scikit-learn	Use Arnio cleaning as a preprocessing layer before model training.
DuckDB / Arrow	Validate and prepare data before analytics and columnar exchange.
notebooks	Inspect quality issues and cleaning suggestions before analysis.

Row-dropping pipeline behavior

Some pipeline steps such as drop_nulls or drop_duplicates can change the number of rows returned during transform.

By default, ArnioCleaner raises a ValueError if a pipeline changes row count during transform because many scikit-learn workflows expect input and output sample counts to remain aligned.

If row-dropping behavior is intentional, pass allow_row_count_change=True when constructing ArnioCleaner.

cleaner = ArnioCleaner(
    steps=[
        ("drop_nulls",),
        ("strip_whitespace",),
    ],
    allow_row_count_change=True,
)

Pandas accessor

df = pd.read_csv("raw_customers.csv")

clean_df = df.arnio.clean(drop_duplicates=True)
quality = clean_df.arnio.profile()
validation = clean_df.arnio.validate({
    "email": ar.Email(nullable=False),
    "user_code": ar.Regex(r"^USR-\d{4}$", nullable=False),
    "age": ar.Int64(nullable=True, min=0),
    "score": ar.Custom("positive"),
})

This keeps pandas as the analysis tool while Arnio handles the preparation, quality, and validation layer.

Product direction: PROJECT_DIRECTION.md

📘 Examples

These examples demonstrate how Arnio integrates with the Python data ecosystem.

They follow a simple workflow:

clean/validate data with Arnio → analyze with other tools

🔹 Interoperability Examples

• Arnio + pandas Clean and normalize messy tabular data using Arnio, then analyze it using pandas. Run:

  python examples/arnio_with_pandas.py

• Arnio + NumPy Prepare numeric data safely using Arnio, then perform computations using NumPy. Run:

  python examples/arnio_with_numpy.py

• Arnio + scikit-learn Prepare messy data with Arnio, then train a model with scikit-learn. Run:

  python examples/arnio_with_sklearn.py

• Arnio + DuckDB Clean data with Arnio, then run SQL queries using DuckDB. Run:

  python examples/arnio_with_duckdb.py

---

🔍 Why Arnio exists

Every data project starts the same way:

df = pd.read_csv("data.csv")              # 💥 RAM spike — entire file as raw strings
df.columns = df.columns.str.strip()        # Why is this not automatic?
df["name"] = df["name"].str.strip()        # Python loop over every cell
df["name"] = df["name"].str.lower()        # Another Python loop
df = df.dropna()                           # Another pass
df = df.drop_duplicates()                  # Another pass

Six lines. Four full-data passes. All in interpreted Python. This is fine for a Jupyter demo — but it doesn't scale, it doesn't compose, and it definitely doesn't belong in production.

Arnio intercepts this entire pattern. It moves the preparation layer into a predictable pipeline, accelerates supported operations in C++, and gives you clean data for pandas, NumPy, scikit-learn, DuckDB, or notebooks.

Without Arnio df = pd.read_csv(path) df.columns = df.columns.str.strip() for col in str_cols: df[col] = df[col].str.strip() df[col] = df[col].str.lower() df = df.dropna(subset=["revenue"]) df = df.drop_duplicates() # 6+ lines, multiple passes, pure Python

With Arnio frame = ar.read_csv(path) df = ar.to_pandas(ar.pipeline(frame, [ ("strip_whitespace",), ("normalize_case", {"case_type": "lower"}), ("drop_nulls", {"subset": ["revenue"]}), ("drop_duplicates",), ])) # Declarative. Single pipeline. C++ execution.

---

🏗️ Architecture

Arnio is not a pandas wrapper. It's a separate runtime with its own data model.

flowchart LR
  subgraph python["Your Python Code"]
    PY["frame = ar.read_csv('data.csv')\nclean = ar.pipeline(frame, [...])\ndf = ar.to_pandas(clean)"]
  end

  python -->|"pybind11 boundary"| cpp

  subgraph cpp["C++ Runtime (_arnio_cpp)"]
    direction TB
    CSV["CsvReader\n• RFC 4180\n• BOM strip\n• Type inference\n• Quoted fields"]
    FRAME["Frame / Column\n• Columnar\n• std::variant\n• Bool null masks\n• O(1) column lookup"]
    CLEAN["Cleaning Engine\n• drop_nulls\n• fill_nulls\n• drop_dupes\n• strip_ws\n• normalize\n• rename/cast"]
    CSV --> FRAME --> CLEAN
  end

  cpp -->|"to_pandas() → zero-copy NumPy buffer (numerics/bools)"| OUT["pandas DataFrame"]

Loading

Design decisions that matter

Decision	What it means
Columnar storage	Data lives in typed std::vectors — vector<int64_t>, vector<double>, vector<string> — not rows of variants. Cache-friendly and SIMD-ready.
Boolean null masks	Nulls are tracked in a separate vector<bool>, keeping data vectors dense. No sentinel values, no NaN tricks.
Two-pass CSV read	Pass 1 infers types across all rows. Pass 2 parses values directly into the correct typed column. No string→object→cast overhead.
Zero-copy bridge	to_pandas() exposes C++ memory directly via NumPy's buffer protocol where supported. Numeric columns preserve the fast zero-copy path by default, while copy=True requests defensive pandas-owned buffers.
Step registry	Pipeline steps map to C++ function pointers. Adding a new cleaning primitive is a single function + one registry entry.

Full architecture documentation: ARCHITECTURE.md API reference guide: Arnio API Reference

---

🏎️ Benchmarks

Reference environment: Ubuntu, Python 3.12, synthetic messy CSV inputs.
Reproduce: make benchmark — generates deterministic tall and wide datasets and runs both engines.

To reproduce the published numbers from a fresh checkout:

python -m venv .venv
source .venv/bin/activate
python -m pip install -U pip
python -m pip install -e .
python benchmarks/generate_data.py
python benchmarks/benchmark_vs_pandas.py

benchmarks/generate_data.py uses deterministic NumPy seeds, so every run creates the same benchmarks/benchmark_1m.csv tall input and benchmarks/benchmark_wide.csv wide input. The benchmark then executes three pandas runs and three arnio runs for each case, printing average wall-clock time from time.perf_counter() and peak Python allocation from tracemalloc. For cleaner comparisons, close other memory-heavy processes and run the script from the repository root after installing the same Python, pandas, NumPy, compiler, and arnio commit you want to compare.

Expected output format:

Tall CSV (1,000,000 rows x 12 columns)
Metric                     pandas        arnio
────────────────────────────────────────────
Exec Time (avg)       4.73s         5.75s
Peak RAM               211MB         212MB
Speed: 0.8x | RAM: -1% reduction

Wide CSV (5,000 rows x 256 columns)
Metric                     pandas        arnio
────────────────────────────────────────────
Exec Time (avg)       ...s          ...s
Peak RAM              ...MB         ...MB
Speed: ...x | RAM: ...% reduction

Small differences are expected across CPUs, operating systems, compilers, Python builds, and pandas/NumPy versions. If you share benchmark results in an issue or PR, include your OS, Python version, CPU model, pandas/NumPy versions, arnio commit, and the full command output so maintainers can compare like for like.

Arnio is near memory parity in the reference benchmark while replacing ad-hoc Python string loops with a compiled, declarative pipeline. Validate memory and speed on your own workload. The execution time gap is a known, active optimization target — the current drop_duplicates and strip_whitespace implementations use unoptimized row-key serialization.

✅ What's already won	🎯 What's being optimized
Native C++ parsing eliminates Python memory spikes Columnar storage matches pandas' internal efficiency Declarative API eliminates .apply() spaghetti Zero-copy bridge for numeric conversions	drop_duplicates — replace string serialization with hash-based comparisons strip_whitespace — in-place mutation instead of copy-on-write Parallel column processing via std::thread Help close the gap →

🧠 Auto Clean Memory Benchmark

To measure the peak memory and execution time of the auto_clean pipeline using realistic dataset sizes:

python benchmarks/benchmark_auto_clean_memory.py --rows 100000

This script generates a reproducible synthetic dataset with mixed column types (strings, ints, floats, booleans, nulls, and duplicates) and measures:

• ar.read_csv performance
• ar.auto_clean(mode="safe") performance (low-risk cleanup like whitespace trimming)
• ar.auto_clean(mode="strict") performance (includes type casting and deduplication)

The dataset is regenerated deterministically unless --reuse-file is provided. Each auto_clean benchmark run reloads the dataset to avoid mutation or caching effects between runs.

Options:

• --repeat N runs each operation multiple times and reports average (and min/max range).
• --seed N changes the deterministic dataset seed.
• --reuse-file reuses an existing dataset file instead of regenerating it.
• --keep-file keeps the generated CSV (otherwise it is removed at the end).

Expected output format:

Operation                    Time(s)     Peak Py(MiB)
--------------------------------------------------------------------
ar.read_csv           0.042 (0.041-0.044)    4.52 (4.50-4.60)
ar.auto_clean(safe)   0.012 (0.011-0.013)    0.15 (0.14-0.16)
ar.auto_clean(strict) 0.035 (0.034-0.036)    1.20 (1.18-1.22)
--------------------------------------------------------------------
Total avg (Read+Strict)       0.077             4.52

---

🧰 Cleaning primitives

Most operations below run natively in C++. Currently, filter_rows, replace_values and standardize_missing_tokens run via the Python (pandas) backend and may be optimized in C++ later.

Primitive	What it does	Example
drop_nulls	Remove rows with null/empty values	ar.drop_nulls(frame, subset=["age"])
drop_columns	Remove selected columns while preserving the remaining order	frame = ar.drop_columns(frame, ["debug_col"])
keep_rows_with_nulls	Keep only rows that contain at least one null	ar.keep_rows_with_nulls(frame, subset=["age"])
validate_columns_exist	Fail early when required columns are missing	ar.validate_columns_exist(frame, ["age"])
filter_rows	Filter rows using comparison operators	ar.filter_rows(frame, column="age", op=">", value=18)
fill_nulls	Replace nulls with a scalar	ar.fill_nulls(frame, 0, subset=["revenue"])
drop_duplicates	Deduplicate rows (first/last/none)	ar.drop_duplicates(frame, keep="first")
drop_constant_columns	Remove columns with only one unique value	ar.drop_constant_columns(frame)
clip_numeric	Clip numeric values to lower and/or upper bounds	ar.clip_numeric(frame, lower=0, upper=100)
coalesce_columns	Select the first non-null value from a list of columns	ar.coalesce_columns(frame, subset=["phone", "mobile"], output_column="contact")
combine_columns	Combine multiple columns into a single output column	ar.combine_columns(frame, subset=["first", "last"], separator=" ", output_column="name")
strip_whitespace	Trim leading/trailing spaces from strings	ar.strip_whitespace(frame)
standardize_missing_tokens	Replace common missing-value strings with NaN	ar.standardize_missing_tokens(frame)
normalize_case	Force lower/upper/title case	ar.normalize_case(frame, case_type="title")
rename_columns	Rename columns via mapping	ar.rename_columns(frame, {"old": "new"})
cast_types	Cast column types	ar.cast_types(frame, {"age": "int64"})
round_numeric_columns	Round numeric columns (non-numeric columns in subset ignored safely)	ar.round_numeric_columns(frame, decimals=2)
replace_values	Replace values using a mapping (column or whole-frame). Handles None/NaN.	ar.replace_values(frame, {"active": "A", "inactive": "I"}, column="status")
clean	Convenience shorthand	ar.clean(frame, drop_nulls=True)
safe_divide_columns	Divide one column by another, handling zero/null denominators	ar.safe_divide_columns(frame, numerator="revenue", denominator="cost", output_column="ratio")
drop_columns_matching	Drop columns whose names match a regex pattern	ar.drop_columns_matching(frame, pattern="^temp_")
trim_column_names	Strip leading/trailing whitespace from column names	ar.trim_column_names(frame)

ArFrame.select_dtypes — type-based column selection

Returns a new ArFrame containing only the columns whose dtype matches the filter. Raises ValueError if no columns match.

frame = ar.read_csv("data.csv")

# Keep only numeric columns
numeric = frame.select_dtypes(include=["int64", "float64"])

# Drop string columns
without_strings = frame.select_dtypes(exclude="string")

Valid dtype strings: "int64", "float64", "string", "bool", "null"

• At least one of include or exclude must be given — raises ValueError otherwise.
• include and exclude must not overlap — raises ValueError if they share a dtype.
• Unknown dtype strings raise ValueError with a list of valid options.
• Raises ValueError when no columns match (never returns an empty frame silently).
• Column order in the result always matches the original frame.

Or compose them all into a pipeline:

clean = ar.pipeline(frame, [
    ("validate_columns_exist", {"columns": ["name", "city", "revenue"]}),
    ("drop_columns", {"columns": ["debug_notes"]}),
    ("strip_whitespace",),
    ("standardize_missing_tokens",),
    ("normalize_case", {"case_type": "lower"}),
    ("fill_nulls", {"value": "unknown", "subset": ["city"]}),
    ("drop_duplicates", {"keep": "first"}),
])

🔁 Replace values

Use replace_values to substitute values using a mapping. It works as a pipeline step (Python backend) and can operate on a single column or the whole frame when column is omitted. It also understands null semantics: using None (or np.nan) as a mapping key targets existing nulls, and mapping a value to None creates real nulls.

Column-specific example:

clean = ar.pipeline(frame, [
    ("replace_values", {"mapping": {"active": "A", "inactive": "I"}, "column": "status"}),
])

Whole-frame example (no column):

clean = ar.pipeline(frame, [
    ("replace_values", {"mapping": {None: "MISSING", "active": "A", "inactive": "I"}}),
])

Direct API:

frame2 = ar.replace_values(frame, {"active": "A", "inactive": "I"})

🔎 Filter rows inside pipelines

Use filter_rows to keep only rows matching a condition.

clean = ar.pipeline(frame, [
    ("filter_rows", {
        "column": "revenue",
        "op": ">=",
        "value": 1000
    }),
])

Supported operators:

• >
• <
• >=
• <=
• ==
• !=

Works with:

• integers
• floats
• strings
• booleans

🔎 Isolate rows with null values

Use keep_rows_with_nulls to audit incomplete data — keep only rows that have at least one null.

frame = ar.read_csv("data.csv")

# Keep all rows that have at least one null anywhere
nulls = ar.keep_rows_with_nulls(frame)

# Keep rows where specifically 'age' or 'score' is null
nulls = ar.keep_rows_with_nulls(frame, subset=["age", "score"])

# Works inside a pipeline too
result = ar.pipeline(frame, [
    ("keep_rows_with_nulls", {"subset": ["age"]}),
])

Useful for data auditing — inspect what's missing before deciding how to fill or drop.

Boolean string normalization

clean = ar.parse_bool_strings(frame)

This normalizes values such as "yes", "no", "true", "false", "y", "n", "1", and "0" into boolean values while preserving unsupported values unchanged.

Columns containing both parsed boolean values and unsupported string values may round-trip as strings because of ArFrame column typing semantics.

### 🔢 Safe column division

Divide one column by another while handling division by zero and null denominators explicitly:

result = ar.safe_divide_columns(
    frame,
    numerator="revenue",
    denominator="cost",
    output_column="ratio",
    fill_value=0.0,  # used when denominator is zero or null
)

When the denominator is zero or null, the result is replaced with fill_value (default 0.0) instead of raising an error or producing NaN/Inf.

---

📊 Pandas Dtype Support Matrix

This table helps users understand which pandas dtypes and workflows are fully supported, partially supported, unsupported, or planned.

If a dtype is partially supported, users may need conversion before processing. Unsupported dtypes should raise clear errors where applicable.

Pandas Dtype	Support Status	Notes / Fix Hints
int64 / Int64	✅ Supported	Fully supported with native C++ columnar storage. Nulls mapped to pd.NA.
float64 / Float64	✅ Supported	Fully supported with zero-copy conversion. Nulls mapped to np.nan or pd.NA.
bool / boolean	✅ Supported	Native booleans supported with C++ backing. Nulls mapped to pd.NA.
string / string[python]	✅ Supported	Native string extension type. Recommended for text. Nulls mapped to pd.NA.
object (strings / scalars)	✅ Supported	Handled as text or coerced to common type if mixed.
object (nested / lists / dicts)	❌ Unsupported	Nested structures not allowed in flat columnar storage. Raises TypeError.
category	❌ Unsupported	Raises TypeError with fix hint. Convert to string: df["col"].astype(str)
datetime64[ns] / timezone-aware	❌ Unsupported	Raises TypeError with fix hint. Use df["col"].astype(str) or string timestamps.
timedelta64[ns]	❌ Unsupported	Raises TypeError with fix hint. Use df["col"].dt.total_seconds().
complex64 / complex128	❌ Unsupported	Raises TypeError with fix hint. Split into real/imag columns or convert to strings.

Notes

• Zero-copy Optimization: Numeric columns (int64, float64) are optimized for fast zero-copy conversion between C++ and pandas where supported.
• Defensive Buffers: Pass copy=True to to_pandas() when downstream pandas code needs defensive pandas-owned column buffers.
• Boolean Buffers: Boolean conversion is copied because std::vector<bool> cannot be exposed as a zero-copy NumPy buffer.
• Null Handling: Columns with null masks are automatically converted to pandas nullable Extension dtypes (Int64, BooleanDtype, StringDtype).
• Index Drop: pandas DataFrame indexes are currently not preserved during from_pandas() conversion; converted frames receive a default RangeIndex when converted back via to_pandas().
• Validation: Attempting to convert any unsupported type will raise a clear, user-friendly TypeError detailing the column name and how to fix/preprocess it.

---

🧠 Data quality engine

Arnio now includes built-in dataset understanding before you analyze in pandas.

report = ar.profile(frame)
print(report.summary())

suggestions = ar.suggest_cleaning(frame)
clean = ar.pipeline(frame, suggestions)

For production data contracts:

# Register a custom validator once, then reference it by name in any schema
ar.register_validator("positive", lambda v: v > 0)

schema = ar.Schema({
    "id": ar.Int64(nullable=False, unique=True),
    "email": ar.Email(nullable=False),
    "phone": ar.PhoneNumber(nullable=False),

    "user_type": ar.String(nullable=False),

    # country becomes required when user_type == "international"
    "country": ar.String(
        nullable=True,
        required_if=("user_type", "international"),
    ),

    # CurrencyCode validates 3-letter uppercase formats (e.g., USD, EUR, INR).
    "currency": ar.CurrencyCode(),

    "username": ar.String(min_length=3, max_length=20),
    "user_code": ar.Regex(r"^USR-\d{4}$", nullable=False),
    "revenue": ar.Custom("positive", nullable=True),
    "signup_date": ar.Date(nullable=False),
    "created_at": ar.DateTime(nullable=False, format="%Y-%m-%d"),

})

result = ar.validate(frame, schema)

if not result.passed:
    summary = result.summary()
    print(summary["issues_by_rule"])
    print(summary["issues_by_column"])
    print(summary["issues_by_column_and_rule"])
    print(result.to_pandas())
    print(result.to_markdown(max_issues=10))

In this example, country becomes required only when user_type == "international".

Date validates strict YYYY-MM-DD calendar dates.

Phone number validation

PhoneNumber() validates common international and formatted phone number strings.

schema = ar.Schema({
    "phone": ar.PhoneNumber(nullable=False),
})

result = ar.validate(frame, schema)
print(result.passed)

Accepted examples include:

• +1-555-123-4567
• +91 9876543210
• 5551234567

Warning-only validation

schema = ar.Schema(
    {
        "age": ar.Int64(
            min=18,
            severity="warning",
        )
    }
)

result = ar.validate(frame, schema)

print(result.passed)  # True
print(result.issue_count)  # Warning issues are still reported

Warning-level issues remain visible in validation results without failing the overall validation status.

Schema JSON round-trips

schema = ar.Schema(
    {
        "id": ar.String(nullable=False),
        "created_at": ar.DateTime(format="%Y-%m-%dT%H:%M:%S"),
    },
    strict=True,
    unique=["id"],
)

payload = schema.to_json()
restored = ar.Schema.from_json(payload)

ValidationResult.to_markdown() is useful in CI logs, GitHub comments, or data quality reports because it renders a compact validation summary plus a GitHub-friendly issue table.

For multi-column uniqueness (composite keys):

schema = ar.Schema({
    "user_id": ar.Int64(nullable=False),
    "course_id": ar.Int64(nullable=False),
}, unique=["user_id", "course_id"])

result = ar.validate(frame, schema)

For low-risk automatic cleanup:

clean, report = ar.auto_clean(frame, mode="strict", return_report=True)

This is the layer pandas does not try to own: profiling, data contracts, row-level validation issues, and safe cleaning suggestions for messy incoming datasets.

Beginner-friendly auto-clean tutorial

Use this workflow when you receive a small messy dataset and want to inspect what Arnio will change before applying it.

import arnio as ar
import pandas as pd

raw = pd.DataFrame(
    {
        "order_id": [1001, 1002, 1002, 1003, 1004],
        "customer": [" Ishan ", " Prasoon ", " Prasoon ", " Pranay ", " Dhruv "],
        "city": [" Paris ", "London", "London", " New York ", " Tokyo "],
    }
)

frame = ar.from_pandas(raw)

report = ar.profile(frame)
summary = report.summary()
print(summary)

suggestions = ar.suggest_cleaning(frame)
print(suggestions)
# [('strip_whitespace', {'subset': ['customer', 'city']}), ('drop_duplicates', {'keep': 'first'})]

safe = ar.auto_clean(frame)
strict = ar.auto_clean(frame, mode="strict")

Messy input:

order_id	customer	city
1001	Ishan	Paris
1002	Prasoon	London
1002	Prasoon	London
1003	Pranay	New York
1004	Dhruv	Tokyo

Expected cleaned output with mode="strict":

order_id	customer	city
1001	Ishan	Paris
1002	Prasoon	London
1003	Pranay	New York
1004	Dhruv	Tokyo

mode="safe" only trims whitespace. Use mode="strict" when you also want deterministic built-in cleanup such as exact duplicate removal.

See examples/auto_clean_tutorial.py for a runnable version of this walkthrough.

For strict mode data-loss risks and safe workflow, see AUTO_CLEAN_GUIDE.md.

Data Quality Reports

Arnio provides detailed profiling for datasets via ar.profile(). To generate the report shown in these examples, the following code was used:

import arnio as ar
import pandas as pd

# Sample dataset used for these examples
data = {
    "user_id": [101, 102, 103, 104],
    "email": ["test@arnio.ai", "invalid-email", None, "test@arnio.ai"],
    "score": [85.5, 90.0, None, 88.2]
}
df = ar.from_pandas(pd.DataFrame(data))
# Bounded profiling for large datasets (controls how many sample values are kept)
report = ar.profile(df, sample_size=5)
safe_report = report.to_dict(redact_sample_values=True)

Profiling privacy and redaction

Profiling helps you understand data, but some report fields can still expose real emails, names, IDs, or other sensitive values. Before you paste output into GitHub issues, Slack, public notebooks, or shared logs, check whether you are sharing aggregate statistics only or raw/sample cell values.

What is aggregate-only vs may expose raw values

Field or export	Aggregate-only?	May expose raw / sample data?
row_count, column_count, duplicate_rows, duplicate_ratio, quality_score, score_components	Yes	No
null_count, null_ratio, unique_count, unique_ratio, whitespace / empty-string counts	Yes	No
Numeric min / max / mean / std / q25–q95	Statistics only	Uncommon on large datasets; small tables can still be identifying
semantic_type, suggested_dtype, warnings	Metadata / hints	Can imply PII type (for example email-like), not redaction
ColumnProfile.sample_values (in-memory)	No	Yes — first N non-null values (sample_size on ar.profile())
ColumnProfile.top_values	Includes counts / ratios	Yes — frequent actual values (exact or approximate; see below)
report.to_dict()	Mixed	Yes — includes sample_values and top_values unless you redact samples
report.to_dict(redact_sample_values=True)	Mixed	sample_values → "[REDACTED]" (same list length); top_values unchanged
report.to_markdown(), report.summary()	Yes	No raw cell values in output
report.to_html() / notebook display of report	Partial	Shows top_values chips; does not list sample_values
report.to_pandas()	Partial	Includes top_values, not sample_values
ProfileComparison.to_dict()	Nested profiles	Yes — embeds left_profile / right_profile via default to_dict()

Arnio does not auto-mask emails, phone numbers, or IDs by column type. Use the controls below for safer sharing.

Safe sharing practices

• JSON logs and artifacts: report.to_dict(redact_sample_values=True) before writing or uploading.
• Collect fewer samples: ar.profile(frame, sample_size=0) skips sample_values (defaults still apply to top_values on string columns).
• Text summaries for CI or comments: prefer report.to_markdown() or report.summary() when you do not need per-value examples.
• Notebooks and HTML exports: avoid evaluating report or saving report.to_html() for sensitive data; HTML still shows top_values.
• GitHub bug reports and examples: use synthetic data (user@example.com, ID-001), a minimal CSV, and redacted to_dict() output — not production dumps.
• Pandas export: ar.to_pandas(frame) returns full table data; redaction applies to quality reports, not the underlying frame.
• Profile comparison: ProfileComparison.to_dict() nests full profiles; build shared artifacts with profile.to_dict(redact_sample_values=True) if needed.

import arnio as ar
import pandas as pd

df = ar.from_pandas(pd.DataFrame({
    "email": ["user@example.com", "bad-email", None],
    "user_id": [101, 102, 103],
}))
report = ar.profile(df, sample_size=2)

# Safer JSON for sharing (sample_values only; top_values still present)
safe_json = report.to_dict(redact_sample_values=True)

# Safer text summary (no sample_values or top_values in output)
print(report.to_markdown())

When approx_top_values=True, high-cardinality string columns estimate top_values from a deterministic sample. Each column may set top_values_is_approximate, top_values_sample_count, and top_values_sample_ratio. Counts and ratios are sample-based, but displayed values are still real strings from your data — treat them like top_values for privacy.

# Optional: approximate top values for high-cardinality string columns
report = ar.profile(
    df,
    approx_top_values=True,
    approx_top_values_min_unique=1000,
    approx_top_values_min_ratio=0.2,
    approx_top_values_sample_size=2000,
)

Notebook dashboard (Jupyter / Colab)

DataQualityReport includes a notebook-friendly HTML dashboard. In a notebook, simply evaluate report in a cell to see a rich, static summary (quality score, duplicates, nulls, warnings, top values, and cleaning suggestions).

If you want to embed or save the HTML explicitly:

from IPython.display import HTML

HTML(report.to_html())
# or: report.to_html(file_path="data_quality_report.html")

Sample output now includes quantiles for numeric columns:

{
  "age": {
    "dtype": "float64",
    "mean": 35.2,
    "std": 10.1,
    "min": 18.0,
    "max": 60.0,
    "q25": 27.5,
    "q50": 35.0,
    "q75": 44.0,
    "q95": 57.0,
    "null_count": 0
  }
}

Compare Profiles

Use ar.compare_profiles() to compare two DataQualityReport profiles and flag per-column drift.

baseline = ar.profile(ar.read_csv("baseline.csv"))
current  = ar.profile(ar.read_csv("current.csv"))

comparison = ar.compare_profiles(baseline, current)
print(comparison.drift_report["score"]["status"])  # "ok", "warning", or "changed"
print(comparison.status_counts)  # {"ok": 2, "warning": 1, "changed": 0}

Use ar.check_quality_gates() when profile drift should become a pass/fail decision for CI, data releases, or monitoring.

result = ar.check_quality_gates(
    baseline,
    current,
    max_row_count_delta_ratio=0.10,
    max_null_ratio_delta=0.05,
    max_numeric_mean_delta_ratio=0.10,
)

if not result.passed:
    print(result.to_markdown())
    result.raise_for_failures()

Scoring Contract: The quality_score starts at 100.0 and subtracts capped penalties for duplicates, nulls, and suggested dtype mismatches. The score_components field exposes these penalties as negative values. (Note: Semantic-validity penalties are intentionally out of scope for the current implementation.)

1. Terminal Representation (Simplified Example)

A simplified view of the standard string representation of the report object:

DataQualityReport(
    row_count=4,
    column_count=3,
    memory_usage=733,
    duplicate_rows=0,
    quality_score=100.0,
    score_components={},
    columns={
        'user_id': ColumnProfile(dtype='int64', semantic_type='identifier', unique_count=4),
        'email': ColumnProfile(dtype='string', semantic_type='categorical', null_count=1, unique_ratio=0.666667, min=13, max=13, mean=13.0),
        'score': ColumnProfile(dtype='float64', semantic_type='numeric', mean=87.9, min=85.5, max=90.0)
    }
)

2. JSON Format (Excerpts from .to_dict())

Key fields from the structured JSON export for integration with APIs or dashboards:

{
  "row_count": 4,
  "column_count": 3,
  "memory_usage": 733,
  "duplicate_rows": 0,
  "duplicate_ratio": 0.0,
  "quality_score": 100.0,
  "score_components": {},
  "columns": {
    "user_id": {
      "dtype": "int64",
      "semantic_type": "identifier",
      "null_count": 0,
      "unique_ratio": 1.0
    },
    "email": {
      "dtype": "string",
      "semantic_type": "categorical",
      "null_count": 1,
      "unique_ratio": 0.666667,
      "min": 13,
      "max": 13,
      "mean": 13.0,
      "warnings": ["contains_nulls"]
    },
    "score": {
      "dtype": "float64",
      "semantic_type": "numeric",
      "null_count": 1,
      "mean": 87.9,
      "min": 85.5,
      "max": 90.0,
      "warnings": ["contains_nulls"],
      "histogram": [
        {"bucket_start": 85.5, "bucket_end": 85.95, "count": 1, "ratio": 0.333333},
        {"bucket_start": 85.95, "bucket_end": 86.4, "count": 0, "ratio": 0.0},
        {"bucket_start": 86.4, "bucket_end": 86.85, "count": 0, "ratio": 0.0},
        {"bucket_start": 86.85, "bucket_end": 87.3, "count": 0, "ratio": 0.0},
        {"bucket_start": 87.3, "bucket_end": 87.75, "count": 0, "ratio": 0.0},
        {"bucket_start": 87.75, "bucket_end": 88.2, "count": 0, "ratio": 0.0},
        {"bucket_start": 88.2, "bucket_end": 88.65, "count": 1, "ratio": 0.333333},
        {"bucket_start": 88.65, "bucket_end": 89.1, "count": 0, "ratio": 0.0},
        {"bucket_start": 89.1, "bucket_end": 89.55, "count": 0, "ratio": 0.0},
        {"bucket_start": 89.55, "bucket_end": 90.0, "count": 1, "ratio": 0.333333}
      ]
    },
    "city": {
      "dtype": "string",
      "semantic_type": "categorical",
      "null_count": 0,
      "top_values": [
        {"value": "London", "count": 3, "ratio": 0.5},
        {"value": "Paris", "count": 2, "ratio": 0.333}
      ]
    }
  },
  "suggestions": [
    {
      "step": "cast_types",
      "kwargs": {"score": "float64"},
      "confidence_score": 0.95,
      "confidence_reason": "Column 'score' conforms perfectly to float64 structure."
    }
  ]
}

3. Example Summary Table

A manually formatted Markdown table representing the core metrics:

Metric	Value
Row Count	4
Column Count	3
Memory Usage	733 bytes
Duplicates	0 (0.0%)
Quality Score	100.0

Bootstrapping a Schema from a Quality Report

After profiling a dataset, you can automatically generate a validation schema directly from the report:

import arnio as ar

frame = ar.from_pandas(df)
report = ar.profile(frame)

schema = ar.Schema.bootstrap_from_report(report)
result = schema.validate(frame)

print(result.passed)
print(result.summary())

The inferred schema uses conservative defaults: column dtypes are mapped directly from the report, and a column is marked nullable=True if any null values were observed during profiling.

🗺️ Roadmap

Version	Focus	Status
v1.0	Stable release · cross-platform wheels · CI/CD · PyPI publishing · Google Colab support	✅ Shipped
v1.1	Production readiness · release hardening · docs/tooling	✅ Shipped
v1.2	C++ pipeline optimization · speed parity with pandas · hash-based deduplication	🔨 Active
v1.3	Chunked / streaming processing · Parquet & JSON readers	📋 Planned
v1.4	Parallel column processing · SIMD string operations	💭 Exploring

For CLI command reference and examples, see CLI_REFERENCE.md.

---

💬 Community

Join the Arnio Discord Community for quick setup help, contributor onboarding, GSSoC 2026 coordination, feature discussion, and community updates.

Discord is for fast conversation and support. GitHub remains the source of truth for issue assignment, PR reviews, bugs, roadmap decisions, and releases.

---

📚 Documentation

• Troubleshooting Guide

🤝 Contribute

Arnio is a GSSoC 2026 project with a structured contributor backlog across beginner, intermediate, and advanced tracks.

You don't need C++ to contribute

Most new features are pure Python pipeline steps:

# 1. Write a function that takes a DataFrame and returns a DataFrame
def remove_special_chars(df, columns=None):
    cols = columns or df.select_dtypes("object").columns
    for col in cols:
        df[col] = df[col].str.replace(r"[^a-zA-Z0-9\s]", "", regex=True)
    return df

# 2. Register it
ar.register_step("remove_special_chars", remove_special_chars)

# 3. Write tests, open a PR. That's it.

If Arnio renames a built-in or registered pipeline step in a future release, the old step name can stay temporarily available and will emit a DeprecationWarning while routing execution to the new canonical step.

If you do know C++

The biggest performance wins are in:

• drop_duplicates — replacing std::ostringstream row serialization with proper hash-based comparisons
• strip_whitespace — converting from copy-on-write to in-place mutation
• Parallel column processing — std::thread across independent columns

Getting started

# macOS / Linux
git clone https://github.com/im-anishraj/arnio.git && cd arnio
make install   # pip install -e ".[dev]" + pre-commit
make test      # pytest with coverage
make lint      # ruff + black

# Windows
pip install -e ".[dev]"
pre-commit install
pytest tests/ -v

PR titles must follow Conventional Commits — feat:, fix:, docs:, chore:. Our release pipeline auto-generates changelogs from these.

For GSSoC contributors, please read GSSOC_GUIDE.md before asking to be assigned. It explains issue claiming, contribution levels, review expectations, and what maintainers look for in a strong PR. If you want a quick onboarding refresher, see the GSSoC FAQ. If you are new to Arnio terms, see the contributor glossary.

📖 Full Contributing Guide ·  GSSoC Guide ·  🐛 Open Issues ·  💬 Discussions ·  Discord

💖 Contributors

Thanks to everyone who contributes to Arnio and helps improve the project.

• View all contributors
• Contribution Guide
• GitHub Discussions

---

🚢 Release process

Arnio releases are automated through Release Please and GitHub Actions.

1. Merge user-facing changes with Conventional Commit PR titles (feat:, fix:, docs:, or chore:) so Release Please can choose the version bump and changelog entries.
2. Review and merge the Release Please PR on main; this updates release metadata and creates the GitHub release and tag.
3. Confirm the Build & Publish Wheels workflow succeeds for the release tag. It builds the sdist and wheels, then publishes to PyPI through Trusted Publishing.
4. Smoke test the published package in a clean environment:

python -m venv /tmp/arnio-smoke
source /tmp/arnio-smoke/bin/activate
python -m pip install -U pip
python -m pip install arnio
printf 'name,revenue\n Ada,10\n' > /tmp/arnio-smoke.csv
python - <<'PY'
import arnio as ar
print(ar.__version__)
print(ar.scan_csv("/tmp/arnio-smoke.csv"))
PY

5. Verify the GitHub release, PyPI project page, and install command all show the expected version before announcing the release.

If any publish or smoke-test step fails, leave the failed tag and GitHub release in place until maintainers agree on the recovery plan.

---

📐 Project structure

arnio/
├── cpp/
│   ├── include/arnio/      # C++ headers — types, column, frame, csv_reader, cleaning
│   └── src/                 # C++ implementations (~30 KB of compiled logic)
├── bindings/
│   └── bind_arnio.cpp       # pybind11 module — the Python↔C++ bridge
├── arnio/
│   ├── __init__.py          # Public API surface
│   ├── io.py                # read_csv, read_jsonl, scan_csv, write_csv, write_parquet
│   ├── cleaning.py          # Python wrappers for C++ cleaning functions
│   ├── pipeline.py          # Step registry + pipeline executor
│   ├── convert.py           # to_pandas (zero-copy), from_pandas
│   ├── frame.py             # ArFrame — lightweight C++ Frame wrapper
│   └── exceptions.py        # ArnioError, UnknownStepError, CsvReadError, TypeCastError
├── tests/                   # pytest suite — CSV, cleaning, pipeline, conversions
├── benchmarks/              # Reproducible arnio vs pandas benchmark
├── examples/                # basic_usage.py, auto_clean_tutorial.py, custom_step.py and ready to run recipes for sales, customers, survey, logs, finance
└── website/                 # Project website — arnio.vercel.app

---

Stop writing cleaning scripts. Declare clean data.

       

_{Built with C++ and pybind11 · Licensed under MIT · Maintained by @im-anishraj}