Filling Data Gaps with the Synthetic Data SDK

Abstract: 

In this hands-on tutorial, we'll explore how to address missing data challenges using synthetic data generation with MOSTLY AI's open-source Synthetic Data SDK. We'll start with an introduction to the SDK, covering both LOCAL and CLIENT modes for training and generating synthetic data.

Then, we'll continue on to advanced techniques such as data rebalancing, smart imputation of missing values, and conditional generation to upsample sparse data areas of interest. These methods help improve dataset completeness, correct imbalances, and boost the performance of downstream models.

Through practical examples and live demonstrations, you'll gain the skills to generate high-quality synthetic data, train GenAI models on enhanced datasets, and integrate synthetic data into your existing workflows.

This session is designed for data scientists, ML engineers, and analytics practitioners across a range of industries - from finance and healthcare to e-commerce and mobility. By the end, you’ll be equipped with actionable techniques to fill data gaps, improve data utility, and make better use of the data you already have.

Session Outline:

Lesson 1: Introduction to the Synthetic Data SDK
Get familiar with the fundamentals of synthetic data and the Synthetic Data SDK. By the end of this lesson, you will be able to install the SDK, run it in LOCAL mode, and generate your first synthetic dataset using a simple HR dataset. You'll understand the difference between LOCAL and CLIENT modes and how to evaluate basic output.

Lesson 2: Smart Imputation of Missing Data
Learn how to handle missing values using synthetic data generation. You will explore how the SDK can be configured to generate complete datasets from incomplete inputs. By the end of this lesson, you’ll be able to fill gaps in your data and compare original vs. imputed values using notebook-based examples.

Lesson 3: Rebalancing Underrepresented Classes
Practice using the SDK to rebalance skewed or imbalanced data. You’ll work with a dataset where certain classes are underrepresented and learn how to adjust generation settings to correct for these imbalances. After this lesson, you’ll be able to produce more balanced datasets for training ML models.

Lesson 4: Conditional Generation for Sparse Segments
Explore conditional generation to target specific slices of your data. You’ll learn how to upsample rare combinations of features by defining generation conditions. By the end of this lesson, you’ll be able to create focused synthetic datasets that improve coverage in critical areas.

After this session, attendees will understand how to use synthetic data to address missing, imbalanced, and sparse data challenges. They will learn how to install and configure the open-source Synthetic Data SDK (https://github.com/mostly-ai/mostlyai), generate synthetic datasets in both LOCAL and CLIENT modes, and apply techniques like smart imputation, rebalancing, and conditional generation. Attendees will gain hands-on experience integrating synthetic data into real-world data science workflows to improve data utility and model performance.

Open source tools used during the session include:
- Python
- Jupyter Notebooks
- Synthetic Data SDK

All code and notebooks will be shared via GitHub for participants to follow along and reuse in their own projects.

Background Knowledge:

This tutorial is designed for data engineers, data scientists, ML engineers, and analysts with basic Python skills and familiarity with working in Jupyter Notebooks. Attendees should have a general understanding of machine learning workflows and working with tabular datasets (e.g., CSV files or pandas DataFrames). No prior experience with synthetic data is required.

To participate fully in the hands-on exercises, attendees should have the following installed before the session: Python 3.11+, Git

Installation and setup of the Synthetic Data SDK will be covered as part of the tutorial. But feel free to get started beforehand by visiting https://github.com/mostly-ai/mostlyai.

Bio: 

Bio Coming Soon!

Open Data Science

 

 

 

Open Data Science
One Broadway
Cambridge, MA 02142
info@odsc.com

Privacy Settings
We use cookies to enhance your experience while using our website. If you are using our Services via a browser you can restrict, block or remove cookies through your web browser settings. We also use content and scripts from third parties that may use tracking technologies. You can selectively provide your consent below to allow such third party embeds. For complete information about the cookies we use, data we collect and how we process them, please check our Privacy Policy
Youtube
Consent to display content from - Youtube
Vimeo
Consent to display content from - Vimeo
Google Maps
Consent to display content from - Google