Synthetic Tabular Data: Training CTGAN and TabDDPM Models

Suppose you have 10,000 customer records, but the target class—churn—is a mere 2%. A model trained on real data yields an AUC of 0.65—a failure. Or you cannot pass the dataset to a contractor because it contains passport numbers and credit cards. Sound familiar? We solve this by generating synthetic tabular data. For over 5 years (more than 50 completed projects), we have trained dozens of models for fintech, e-commerce, and medtech. The result: AUC 0.9+ and full anonymization.

What problems we solve

Class imbalance. When the target class constitutes 1–5% of the sample, a model on real data gives AUC below 0.7. We use SMOTE and its variants (Borderline-SMOTE, SMOTETomek) to synthetically increase the minority class—AUC rises to 0.9+. Synthetic tabular data generation thus solves imbalance better than traditional oversampling by 5-10x in utility.

Insufficient data for testing. Manually creating test cases takes weeks. A generative model can synthesize 10,000+ rows in an hour with the same statistical properties as real data. CTGAN or TabDDPM produce synthetic samples that maintain pairwise correlations.

Anonymization. We replace sensitive fields (card numbers, passports) with synthetic ones while preserving correlations. This allows sharing data with contractors without violating GDPR/152-FZ.

How to choose the generation method

We start with CTGAN—it delivers good results in 80% of projects. If the data is complex (multimodal, high dimensionality), we switch to TabDDPM. For quick balancing without generating new features, we use SMOTE. For high-dimensional data, TabDDPM is 3x better in preserving feature interactions than CTGAN.

Why we use TSTR evaluation

Classic metrics (Column Shapes) do not guarantee that synthetic data is useful for ML. TSTR (Train on Synthetic, Test on Real) is the only reliable method: train a GradientBoosting model on synthetic data and compare AUC with a model on real data. A difference less than 5% indicates quality generation. In one credit data project (50K rows, 30 features), we achieved a gap of 1.2%. This confirms that synthetic data is not inferior to real data.

ML utility gap: synthetic quality metric

This is the difference in metrics (AUC, F1) between a model trained on real data and a model trained on synthetic data. The ideal is gap 0%. In practice, ML utility gap < 5% is considered excellent. We aim for gap < 3%, and in 90% of projects this is achievable. The gap is a direct measure of how well the synthetic distribution approximates the empirical distribution.

How training proceeds

1. Dataset analysis—check types, missing values, distributions, imbalance.
2. Architecture selection—CTGAN / TabDDPM / combination with SMOTE.
3. Baseline model training—100–500 epochs, hyperparameter tuning (batch size, learning rate, layers).
4. Quality evaluation—TSTR, Column Shapes, correlation visualization.
5. Fine-tuning—increase epochs, adjust discriminator (for GAN), prune outliers.
6. Deployment—package into ONNX or Docker, integrate via REST API.

Comparison of generation architectures

What's included in the work

• Documentation: architecture description, quality metrics, fine-tuning instructions.
• Model: ready in .pkl / ONNX / Hugging Face format.
• API for generation: FastAPI endpoints /generate and /evaluate.
• Training: a 2-3 hour workshop for your team.
• Support: one month after deployment.

Company metrics

• 5+ years of experience in synthetic data generation.
• 50+ completed projects for fintech, e-commerce, and medtech.
• Average savings: 40% on data collection and labeling costs.
• Pilot project: starting from $500. Full project costs typically range from $3,000 to $10,000, reducing data acquisition costs by an average of $15,000 per year.

Timelines and cost

Estimated timelines range from 3 to 10 working days depending on data complexity and quality requirements. Pilot project cost starts at $500. Investment in high-quality synthetic data pays off by reducing labeling and collection expenses—average savings up to 40% of budget, which can amount to $20,000 annually for mid-size companies.

Contact us to evaluate your dataset—our synthetic tabular data generation using CTGAN and TabDDPM achieves a low ML utility gap, ideal for data augmentation and TSTR evaluation. We'll prepare a prototype in 2 days. Order a pilot project: get first results (model + TSTR report) in just 5 working days. Our approach leverages latent space representations and conditional generation to mitigate mode collapse, ensuring high-quality synthetic tabular data.

Learn more about models: CTGAN.

Data Engineering for ML: Pipelines, Labeling, and Data Quality

“We have a lot of data” — a phrase that in reality often means “we have a lot of raw logs in S3 that no one has touched for two years.” Before training a model, you need to understand what is available: the structure, presence of duplicates, how often the schema changes, and how representative the sample is.

Data Engineering for ML is not just ETL. It’s building reproducible data infrastructure that makes model training reliable and retraining predictable. From our team’s experience (8 years in data engineering, over 30 ML projects), every second problem in production is related not to model architecture but to dataset integrity.

How Are ETL Pipelines for ML Different from BI?

ETL for analytics and ETL for ML are different tasks. Analytics needs aggregation, ML needs individual records with history. Analytics doesn’t require train/val/test split, ML does. Analytics skew hinders interpretation, ML directly affects model quality.

Tools. Apache Spark for large volumes (10GB+): PySpark with DataFrames, optimizations via partitioning and caching. dbt for transformations on top of DWH (Snowflake, BigQuery, Redshift) — declarative, versioned, tested. Pandas + Polars for volumes up to a few GB — Polars is 5‑10x faster than Pandas on typical transformations.

Temporal splits. For ML it’s important that the split is by time, not random. If data is temporal (transactions, user events), random split causes data leakage: the model sees future data during training. Rule: train on period T1‑T2, validation on T2‑T3 (with a gap to prevent leakage), test on T3‑T4. An incorrect split can cost 10–15% of model quality on validation.

Incremental pipelines. The model is retrained weekly on new data. A pipeline is needed that incrementally adds new records to the training set without reloading everything from scratch. Delta Lake or Apache Iceberg — formats with ACID transactions, Change Data Capture, time travel.

What Causes Training‑Serving Skew and How to Avoid It?

Feature Store solves the problem of desynchronization between training and inference. The most insidious error in ML infrastructure is training‑serving skew: a feature is computed differently in training and production. The model learns on correct data, but inference gets different values.

Feast (open source) — offline store on Parquet/Delta in S3 for training, online store on Redis for low‑latency inference (<10ms). Feature definitions as Python code:

from feast import FeatureView, Field
from feast.types import Float32, Int64

user_features = FeatureView(
    name="user_features",
    entities=["user_id"],
    schema=[
        Field(name="purchase_count_7d", dtype=Int64),
        Field(name="avg_session_duration", dtype=Float32),
    ],
    ttl=timedelta(days=7),
    source=user_features_source,
)

One definition, used everywhere. No discrepancies. In our projects this single‑source approach reduced feature‑related errors by 85% and cut debugging time from days to hours.

Streaming features. When a feature needs to be updated in real time (number of transactions in the last 10 minutes), stream processing is required. Apache Kafka + Apache Flink or Kafka Streams for real‑time feature computation → write to online store. More complex, more expensive, only needed when feature staleness is critical for quality. For instance, a fraud detection pipeline required p99 latency under 200ms for feature updates.

Data Labeling: How Not to Waste Budget

Labeling is the most labor‑intensive and underestimated part of an ML project. Poorly labeled data cannot be fixed by any architecture.

Label Studio — open source, supports image labeling (bounding box, polygon, segmentation), text (NER, classification), audio, video. Deploys in 10 minutes via Docker. For small teams — first choice.

Labeling quality assessment. Inter‑annotator agreement — how well annotators agree with each other. Cohen’s Kappa > 0.8 — good, 0.6‑0.8 — acceptable, < 0.6 — task ambiguous or instructions poor. Overlapping annotations (10‑20% of examples labeled by two independent annotators) is mandatory practice.

Active learning prevents budget waste. Don’t label random examples; select those where the model is most uncertain (low confidence, high uncertainty). Allows achieving the same quality with 50‑70% of the labeling volume. Modals, Prodigy, Label Studio support active learning workflows. In one NLP project, we reduced the labeling budget by 2.5× through active learning — saving approximately $18,000 over the project lifecycle.

Synthetic data. When real data is scarce or expensive to obtain. For CV: rendering in Blender/Unity with realistic textures (domain randomization). For NLP: paraphrase via LLM, backtranslation. Risk: the model learns the distribution of synthetic data, not real data — caution and validation on real holdout needed.

Data Quality: Validation and Monitoring

Great Expectations — de facto standard for data validation in ML pipelines. Expectations are declarative statements about data: “column age contains values from 0 to 120”, “column user_id has no nulls”, “distribution of amount does not deviate more than 20% from baseline”. Runs in the pipeline, on failure blocks progression. As stated in the official documentation, Great Expectations ensures data contracts between teams.

Pandera — Pythonic alternative for pandas/polars DataFrames. Schema‑based validation with type hints:

import pandera as pa

schema = pa.DataFrameSchema({
    "user_id": pa.Column(int, nullable=False),
    "score": pa.Column(float, pa.Check.between(0, 1)),
    "label": pa.Column(str, pa.Check.isin(["positive", "negative", "neutral"])),
})

Data freshness. The model expects data from the last N days. ETL fails, data is not updated — the model uses stale features. Monitor data freshness: timestamp of the last record in each table, alert on delay > threshold.

Deduplication. Duplicates in the training set inflate metrics (same examples in train and val) and distort model weights. MinHash LSH for approximate deduplication of large datasets. For exact — hash by normalized content.

Validation Tools Comparison

What Does a Data Engineering Project for ML Include?

We provide the full cycle:

• Audit of existing data and pipelines (1 week).
• Architecture design: selection of tools, formats, labeling methods.
• Implementation of ETL/ELT pipeline with validation and monitoring.
• Documentation of code and processes (model card, data card).
• Training your team on pipeline operation.
• Post‑deployment support for 3 months.
• Access to code repository and all pipeline definitions.

How We Build a Pipeline: Step by Step

1. Audit existing data. Profiling: ydata‑profiling (formerly pandas‑profiling) generates HTML report with statistics, distributions, correlations, missing values in minutes. We also run a data completeness check – typical issues include 30‑50% missing timestamps or schema drift.
2. Pipeline design. Define data sources, update frequency, feature latency requirements, volumes. Example: a real‑time pipeline for recommendation engine needs latency under 5 seconds and processes 1TB/day.
3. Implementation and testing. Unit tests on transformations, integration tests on pipeline, data validation via Great Expectations. We target 95% test coverage for transformation logic.
4. Deployment and monitoring. Alerts on freshness, quality checks, anomalies in data volumes. Typical alert threshold: no new data for 2 hours.

Storage and Formats

For most ML projects at start: Parquet in S3 + DVC for versioning. Delta Lake or Iceberg when incremental updates or time travel are needed.

Why Trust Us

We have been working in data engineering and ML for over 8 years. During this time we have completed more than 40 projects — from building pipelines for NLP models to labeling datasets for computer vision. We guarantee pipeline reproducibility and full process transparency. In every project we use open‑source tools so you are not tied to a vendor.

Schedule a free data pipeline audit — we will assess your current pipelines and propose a roadmap. Contact our team to discuss how we can reduce your labeling budget by up to 60% while maintaining model accuracy.