---
id: fasttransform
version: "0.0.2"
license: Apache Software License 2.0
license_treatment: permissive
maintenance: active
---

# fasttransform — Transform is the main building block of data pipelines in fastai. And elsewhere if you want.

License: permissive · Maintenance: active · Downloads: 216.0K/mo

## What it is and what it does

fasttransform is a data transformation framework that wraps functions into Transform objects, enabling them to be composed into reusable pipelines. The core idea is that a single Transform can encode data forward and decode it backward, specialize behavior based on input types, and preserve or convert types intelligently. It depends on fastcore for core utilities and plum-dispatch for type-based function selection.

The package is designed for data pipeline construction where you need transformations that are reversible (useful for normalization/denormalization), type-aware (handling multiple input formats), and composable into larger workflows. It supports decorator syntax for simple cases and class-based definition for stateful transforms that need setup phases, like computing statistics from a dataset before normalizing.

Use it for:

- Build reversible normalization transforms that can both encode data (e.g., z-score) and decode predictions back to original scale.
- Handle heterogeneous input types in a pipeline by defining type-specialized encode functions that dispatch automatically.
- Compose multiple transforms into a Pipeline for end-to-end data processing with single encode/decode calls.
- Create stateful transforms that compute aggregate properties (mean, std) during setup and apply them consistently across data.
- Preserve runtime subtypes through transformations so that custom numeric types remain their original type after operations.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

fasttransform provides a Transform class for building reusable, composable data transformations with support for reversibility, type-based dispatch, and type preservation.

Yes, if you are building data pipelines that need reversible, type-aware transformations. The low install friction, active maintenance, permissive license, and clean API make it a reasonable choice. However, verify that it fits your specific dispatch and composition needs—it is still in Beta (version 0.0.2) and may not yet be production-hardened for all use cases.

## Install

pip install fasttransform
uv add fasttransform
poetry add fasttransform

## Installing fasttransform

Before you install: Low install friction with only two runtime dependencies (fastcore and plum-dispatch). The package is actively maintained with a recent commit on 2026-07-26 and has been in active development since its first release.

License in practice: Licensed under Apache Software License 2.0, a permissive license that allows commercial use, modification, and distribution with minimal restrictions.

Quickstart:

from fasttransform import Transform, Pipeline

@Transform
def add_one(x):
    return x + 1

result = add_one(2)  # returns 3

# Reversible transform
def enc(x): return x*2
def dec(x): return x//2
t = Transform(enc, dec)
encoded = t(2)  # returns 4
decoded = t.decode(encoded)  # returns 2

Verify before relying:

- Whether type-based dispatch via plum-dispatch handles edge cases or has known limitations with complex type hierarchies.
- Performance characteristics when chaining many transforms in a pipeline with large datasets.
- Whether the package is actively used in production fastai workflows or remains primarily experimental.

## Package facts

- License: Apache Software License 2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 216.0K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

data transformation pipeline, reversible data transforms, type-based function dispatch, data encoding decoding, reusable data pipeline, type conversion preservation, function composition framework, data-pipeline, type-dispatch, reversible-transform

[View on SkillFed](https://skillfed.io/packages/fasttransform) · [View on PyPI](https://pypi.org/project/fasttransform/)