RDT (Reversible Data Transforms)

1.21.0 · active · verified Wed Apr 15

RDT (Reversible Data Transforms) is a Python library that enables the transformation of raw data into fully numerical data, making it ready for various data science tasks. The transformations are designed to be reversible, allowing conversion back to the original data format. It is part of The Synthetic Data Vault Project and is actively maintained by DataCebo, with frequent updates and releases. The current version is 1.21.0.

Warnings

breaking RDT versions prior to 0.2.0 had a significantly different API. Version 0.2.0 introduced a brand new API, removed the old metadata JSON from user arguments, and made transformers work exclusively with pandas Series.
Fix: Upgrade to RDT 0.2.0 or newer and refactor code to use the new API, especially `HyperTransformer` and its configuration. Ensure data is in pandas DataFrame format.
breaking Version 0.6.0 brought major changes to the `HyperTransformer` and `BaseTransformer` APIs, enabling multi-column input for transformers and allowing sequences of transformers per column.
Fix: Update code to align with the new `HyperTransformer` and `BaseTransformer` APIs introduced in 0.6.0, leveraging the enhanced multi-column and chained transformer capabilities.
deprecated The `frequencyEncoder` transformer is deprecated and will not be supported in future RDT versions.
Fix: Migrate to using the `UniformEncoder` transformer instead.
deprecated The distribution option names for `GaussianNormalizer` have been updated to be consistent with `scipy`. `gaussian` is now `norm`, `student_t` is `t`, and `truncated_gaussian` is `truncnorm`.
Fix: Update distribution names in `GaussianNormalizer` configurations to the new `scipy`-consistent terms.
gotcha The `sdtype` 'text' was removed in RDT versions 1.13.0 and newer. Attempting to use 'text' as an sdtype will lead to errors.
Fix: Replace the 'text' sdtype with 'id' for relevant columns.
gotcha Python 3.6 support was dropped in RDT 1.0.0, and later versions have stricter Python requirements (e.g., currently >=3.9, <3.15).
Fix: Ensure your Python environment meets the required version for your RDT installation. For version 1.21.0, Python 3.9 through 3.14 are supported.

Install

pip install rdt Install with pip

Imports

HyperTransformer
```
from rdt import HyperTransformer
```
Used for transforming multi-column datasets.
get_demo
```
from rdt import get_demo
```
Provides a demo dataset for quick experimentation.

Quickstart

This quickstart demonstrates how to load a demo dataset, initialize a `HyperTransformer`, automatically detect its configuration based on the data, transform the data into a numerical format, and then reverse the transformation back to its original representation.

import pandas as pd
from rdt import HyperTransformer, get_demo

# Load a demo dataset
customers = get_demo()
print("Original Data:\n", customers.head())

# Initialize and detect config with HyperTransformer
ht = HyperTransformer()
ht.detect_initial_config(data=customers)
print("\nDetected Config:\n", ht.get_config())

# Transform the data
transformed_data = ht.transform(customers)
print("\nTransformed Data (first 5 rows):\n", transformed_data.head())

# Reverse transform the data back to original format
reversed_data = ht.reverse_transform(transformed_data)
print("\nReversed Data (first 5 rows):\n", reversed_data.head())

view raw JSON →