Today, the DataCebo team is happy to announce a new major version of our software, SDV Enterprise 2.0.
What’s Changed?
You’ll notice that a large majority of the API is exactly the same. Our goal was to keep everything that’s been working and introduce new or breaking changes only when we felt that the benefit would be worth it. These release notes cover the major features you’ll now see in SDV 2.0, aimed to streamline your synthetic data creation workflow.
For more details about our vision and how our team is thinking about the synthetic data space, check out our launch blog post here.
Specify Ordinal Variables
When using SDV, you specify data types using the sdtypes designation. You can choose from a variety of data types representing PII values (such as phone numbers and addresses) to statistical concepts (such as numerical and categorical columns). In SDV 1.0, the “categorical” designation used to cover both unordered (nominal) and ordered (ordinal) cases, though it required extra setup for the order.
Starting from SDV 2.0 you can designate columns as ordinal (ordered categorical) directly within the metadata as well as specifying thie order. Use this for discrete data that contains an inherent order, such as scaled survey response scores (strongly_disagree, disagree, agree, etc.) as well as encoded variables that take on discrete values (1, 2, or 3). SDV will automatically applies the correct pre-processing for discrete, ordinal variables.
"survey_response": {
"sdtype": "ordinal",
"range_values": ["strongly_disagree", "disagree", "neither_agree_nor_disagree",
"agree", "strongly_agree"]
}
Enhanced Metadata with Ranges
SDV 2.0 offers enhanced metadata that allows you to input the full ranges that you expect from the data. This includes the min/max ranges for categorical columns, full category values for categorical data, ordered category values for ordinal data, whether null values are allowed, decimal precision, and more.
"age": {
"sdtype": "numerical",
"range_min": 18,
"range_max": 120,
"range_is_nullable": true,
}
Adding this to the metadata allows SDV to directly use this information to create higher quality data:
- Your training data might contain only a subset of the allowable range of values. For example, the age range in your training data could contain adults aged 23-67 but the fully allowable range is broader at 18+. In this case, the Targeted Sampling bundle will automatically extrapolate ranges for you based on the metadata, and ensure that you can create the synthetic data within the fully allowable ranges.
- If you’re looking to create synthetic data from scratch, the DayZSynthesizer will automatically adapt to the ranges and values that are contained in the metadata. No extra setup needed.
- When evaluating your data, the Diagnostic Report automatically checks that the synthetic data follows the provided ranges in the metadata for validity.
Consistent Input/Output
In SDV 2.0, all synthesizer will accept and return data in the same format: Provide your as a dictionary mapping the table name to a pandas DataFrame containing the data. And as a result, the synthesizers will always provide you synthetic data in that same format. No need to switch between formats for single-table vs. multi-table use cases.
training_data = {
'users': users_dataframe,
'transactions': transactions_dataframe
}
The demo datasets, modeling, sampling, and evaluation now follow the same signatures for their methods.
Upgrade Timeline
Starting from today, follow the usual Installation Instructions to install and use SDV 2.0. Any new features will be added to SDV 2.0 from now on.
If your team still requires some time to upgrade your systems, SDV Enterprise 0.53 will still be available for 6 months time. Just pass in the --version flag to install a specific, older version of SDV Enterprise.
sdv-installer install --username user@company.com --license-key 123 --version 0.53.0
This method will work for 6 months, during which we’ll provide light support and bug fixes only. After the 6 month period, we will only start supporting SDV Enterprise 2.0.
Resources
For more information about using SDV and learning about synthetic data, check out the resources below. And if you have any questions, please feel free to start a new post in the Forum.
Happy synthesizing!
Release History
2.0.0 (Oct 16, 2026): The initial release of v2.0 release with the new features for ordinal variables, metadata enhancements, and consistent input/output.