DataZip

DataZip is a Python library that extends zipfile.ZipFile to provide seamless serialization and deserialization of complex Python objects — a more flexible and readable alternative to pickle for data science workflows.

Why DataZip?

• Human-inspectable archives: DataZip files are standard .zip files. You can open them with any archive tool and inspect the contents.
• Broad type support: Works out of the box with pandas DataFrames/Series, NumPy arrays, Polars DataFrames, datetimes, paths, sets, frozensets, complex numbers, and custom classes.
• Efficient storage: Tabular data is stored as Parquet; arrays as .npy. JSON is used for metadata and simple types.
• No pickle by default: Most types are serialized without pickle, making files safer and more portable.
• Custom class integration: Any class that implements __getstate__/__setstate__ (the standard pickle protocol) works automatically. The IOMixin makes it even simpler.

Quick Example

from io import BytesIO
import pandas as pd
from datazip import DataZip

# Write
buffer = BytesIO()
with DataZip(buffer, "w") as z:
    z["df"] = pd.DataFrame({"x": [1, 2, 3], "y": [4, 5, 6]})
    z["config"] = {"threshold": 0.5, "labels": ["a", "b"]}
    z["values"] = {1, 2, frozenset([3, 4])}

# Read
with DataZip(buffer, "r") as z:
    df = z["df"]
    config = z["config"]

Supported Types

Category	Types
Primitives	str, int, float, bool, None, complex
Collections	dict, list, tuple, set, frozenset, deque, defaultdict
Date/Time	datetime, pandas.Timestamp
Paths	pathlib.Path
Custom	Any class with __getstate__/__setstate__
Optional	numpy.ndarray, pandas.DataFrame, pandas.Series, polars.DataFrame, polars.LazyFrame, polars.Series, Plotly figures

Installation

See the Installation page for full details including optional dependencies.