dataclassframe

A dataclass container with multi-indexing and bulk operations. Provides the typed benefits and ergonomics of dataclasses while having the efficiency of Pandas dataframes.

The container is based on data-oriented design by optimising the memory layout of the stored data, providing fast bulk operations and a smaller memory footprint for large collections. Bulk operations are enabled using Pandas which has a rich set of vectorised methods for both numerical and string data types.

Multi-indexing provides the ability to use multiple fields as keys to index the records. This is suitable for bidirectional and inverse dictionary keys.

A DataClassFrame provides good ergonomics for production code as columns are immutable and columns/data types are well defined by the dataclasses. This makes it easier for users to understand the "shape" of the data in large projects and refactor when necessary.

Installing

Get the latest version using pip/PyPi

pip install dataclassframe

Feature comparison

*DataFrames used in production code as an data interchange format is considered by the author and others as an anti-pattern.

Show by example

A container data-type for dataclasses...

from dataclasses import dataclass
from dataclassframe import DataClassFrame

@dataclass
class ExampleDC:
    field1: str
    field2: int

records = [
    ExampleDC('a', 1),
    ExampleDC('b', 2),
    ExampleDC('c', 3),
]

dcf = DataClassFrame(
    record_class=ExampleDC,
    data=records,
    index=['field1', 'field2']
)

Which acts like a ordered dictionary with multi-indexing...

# Obtain record `ExampleDC('b', 2)`
row_idx = dcf.iat[1]    # Using positional index
row_f1 = dcf.at['b']    # Using index of `field1`
row_f2 = dcf.at[:, 2]   # Using index of `field2`
assert row_idx == row_f1 == row_f2

With bulk operations on the columns..

assert dcf.cols.field2.sum() == 6

Works nicely with Python 3 type hints...

dcf: DataClassFrame[ExampleDC]
dcf.iat[1]: ExampleDC

Design

It's no secret that under the hood DataClassFrames are using Pandas DataFrames to store data. The data is converted where possible to Pandas Series, which in turn use Numpy arrays. When the user accesses a record the data is then converted back into the dataclass provided at initialisation.

Pandas provides many advantages over of using a simple list of dataclasses such as better memory footprint and fast vectorised operations. Each column of data is stored (usually) using Numpy arrays which use a continuous block of memory. Providing faster access and better CPU cache characteristics.

However using Pandas DataFrames directly in production code as a data interchange format is considered by the author and others as an anti-pattern. Specifically as DataFrames are column-wise mutable and therefore difficult to determine at code-time what columns the dataframe contains i.e. its shape. Users of DataFrames will typically add new columns to add new features.

DataClasses on the other hand are attribute-wise immutable. Therefore the intentions and shape of the data are clear, making it easier to refactor and maintain the code.

DataFrames also do not provide any type-hinting benefits, while DataClass do as they are attribute-wise immutable.

DataClassFrames provide the benefits of both worlds by defining the data shape upfront by using a DataClasses. The columns and record data types are defined by the provided dataclass.

Todo

• Slicing and dataclassframe views for accessing data and setting data
• Append and inserts
• Data-oriented design for Numpy fields

Changelog

All notable changes to this project will be documented here.

The format is based on Keep a Changelog and this project adheres to Semantic Versioning.

[0.1.0] - 2020-10-22

Added

• Initial release of dataclassframe

License

© Josh Levy-Kramer 2020. dataclassframe is released under the MIT license.