Synthetic Tabular Data for ML: Generation, Code, Validation

Collecting real data for ML often runs into compliance constraints: GDPR, HIPAA, or corporate policies prohibit transferring raw datasets to developers. In a project with bank transactions, we couldn't use real records — we had to generate synthetic data preserving distributions and correlations up to KS p-value > 0.4. Synthetic data solves this problem: expands datasets, balances classes (e.g., turn 1% fraudulent transactions into 50%), tests models without breach risk. In our work, we have implemented generation for 15+ projects in finance, healthcare, and retail. Each project requires an individual approach — no universal solution exists. The problem is especially acute with imbalanced datasets: if the target class is less than 1%, without synthetic data you cannot train a model. We use CTGAN with conditional sampling to generate examples of the required class in the right proportion. Savings on data collection — up to 40%, and on compliance audit — up to 30%.

Why is synthetic tabular data necessary for ML?

The key pain point is the shortage of high-quality labeled data. Even if data exists, it often contains personal information inaccessible to external teams. Synthetic data removes these restrictions: you get a dataset with the same statistical properties but without leakage risk. For imbalanced tasks (fraud detection, rare diseases), this is the only way to obtain a representative sample of the minority class. Savings on data collection can reach 40%, and development speed increases manyfold.

How synthetic tabular data solves the problem of imbalanced classes

CTGAN uses a conditional vector that sets the desired ratio of categories in the generated dataset. For example, for fraud detection we fix the share of fraudulent transactions at 50% — this sharply improves the model's recall. The generator and discriminator compete: the former learns to create realistic records, the latter to distinguish them from real ones. As a result, distributions and correlations are preserved with high accuracy. For financial data, typical KS p-value for numerical features is 0.2–0.6.

Choosing a generation method — synthetic tabular data

The choice depends on the data type and privacy requirements. Based on our experience:

Gaussian Copula works 10 times faster than CTGAN, but CTGAN better preserves complex multimodal distributions. For imbalanced classes, CTGAN guarantees exact class ratio after generation via conditional vector. We tune hyperparameters (embedding_dim, generator_dim) for each dataset — GPU utilization reaches 90% per epoch.

How to improve synthetic data quality with fine-tuning

Fine-tuning a generative model on a specific domain improves quality. For medical data, we fine-tune a pretrained CTGAN for 10 epochs with a reduced learning rate. Result: KS p-value improves from 0.05 to 0.4. However, it's important not to overfit — we use early stopping based on SDMetrics metrics. For each project, we create a model card that records hyperparameters, metrics, and generation conditions.

How to evaluate the quality of synthetic data

Validation is a key stage. We use SDMetrics and scipy to check distributions and correlations. The goal is for synthetic data to be statistically indistinguishable from real (p-value > 0.05).

from scipy.stats import ks_2samp
import matplotlib.pyplot as plt

def validate_synthetic_quality(real: pd.DataFrame, synthetic: pd.DataFrame) -> dict:
    results = {}
    for col in real.select_dtypes(include=np.number).columns:
        ks_stat, p_value = ks_2samp(real[col].dropna(), synthetic[col].dropna())
        results[col] = {
            'real_mean': real[col].mean(),
            'synthetic_mean': synthetic[col].mean(),
            'real_std': real[col].std(),
            'synthetic_std': synthetic[col].std(),
            'ks_stat': ks_stat,
            'distribution_match': p_value > 0.05
        }
    real_corr = real.select_dtypes(np.number).corr()
    synth_corr = synthetic.select_dtypes(np.number).corr()
    corr_diff = (real_corr - synth_corr).abs().mean().mean()
    results['correlation_mae'] = corr_diff
    return results

Typical quality thresholds achievable in practice:

For most ML tasks, synthetic data generated by CTGAN with a score > 0.85 on SDMetrics allows achieving 95–98% of model quality compared to training on real data of the same volume.

import pandas as pd
from ctgan import CTGAN
import numpy as np

def train_ctgan_synthesizer(
    data: pd.DataFrame,
    discrete_columns: list,
    epochs: int = 300
) -> CTGAN:
    synthesizer = CTGAN(
        embedding_dim=128,
        generator_dim=(256, 256),
        discriminator_dim=(256, 256),
        batch_size=500,
        epochs=epochs,
        verbose=True,
        pac=10,
    )
    synthesizer.fit(data, discrete_columns=discrete_columns)
    return synthesizer

financial_data = pd.read_parquet("transactions.parquet")
discrete_cols = ['merchant_category', 'transaction_type', 'currency', 'is_fraud']
synth = train_ctgan_synthesizer(financial_data, discrete_cols)
n_real = len(financial_data)
synthetic = synth.sample(n_real * 5)
print(f"Real fraud rate: {financial_data['is_fraud'].mean():.4f}")
print(f"Synthetic fraud rate: {synthetic['is_fraud'].mean():.4f}")

When validation is critical: typical mistakes

• Overfitting the generator: if the model memorized real rows, KS p-value is anomalously high (> 0.9). Check for duplicates.
• Wrong metric choice: KS test alone is insufficient — be sure to look at correlation MAE and category coverage.
• Ignoring data types: categorical features with rare values require increased batch_size and epochs.

What is included in the development of a synthetic data pipeline?

We offer a full cycle of work:

1. Dataset analysis: evaluation of distributions, correlations, missing values, and outliers.
2. Model selection and tuning: choosing architecture (CTGAN, TVAE, Gaussian Copula) and hyperparameters.
3. Training and validation: using SDMetrics, KS tests, correlation analysis.
4. Integration: creating a pipeline on Airflow or Docker, API for generation.
5. Documentation: model card with metrics, operation manual, quality report.
6. Team training: workshop on using the synthesizer and interpreting metrics.
7. Warranty support: 2 weeks after implementation.

Timelines — from 2 to 4 weeks depending on dataset complexity. Cost is calculated individually. Get a consultation on choosing a generation method for your dataset — contact us. Order development of a pipeline for your data.

MLOps: Infrastructure for Training, Deploying, and Monitoring ML Models

The model is trained, metrics — F1 0.94 on validation. Three months later in production, quality drops by 12%. No one knows when — there is no monitoring. It's impossible to retrain quickly — the training script is in a Jupyter notebook of a data scientist who has already left. Data for retraining is collected manually from three disparate systems. About half of the projects come to us with this pain. We build a turnkey MLOps platform: from experiment tracking to automatic deployment and data drift monitoring. We will assess your infrastructure in 1–2 weeks, and in 4–6 weeks you will get a basic MLOps core running in production. Our team has 10+ years of experience in ML infrastructure, over 50 implementations.

How does MLOps infrastructure benefit your ML projects?

Experiment Tracking and Reproducibility

Without tracking, an ML project turns into chaos: it's unclear which checkpoint is better, which hyperparameters were used, which dataset. Reproducing a result a month later is a quest.

Why is experiment tracking the foundation of reproducibility?

MLflow is an open source standard for tracking. It logs parameters, metrics, artifacts (models, graphs), and code. MLflow Model Registry is a centralized model storage with versioning and lifecycle stages (Staging → Production → Archived). Deployment via MLflow Serving or integration with external systems.

Typical initialization in code:

import mlflow

mlflow.set_experiment("fraud-detection-v2")
with mlflow.start_run():
    mlflow.log_params({"learning_rate": 3e-4, "batch_size": 64, "epochs": 10})
    mlflow.log_metric("val_f1", val_f1, step=epoch)
    mlflow.pytorch.log_model(model, "model")

This is the minimum. In production, we add logging of system metrics (GPU utilization, memory), dataset (hash, version), code (git commit hash). Weights & Biases — richer UI, collaboration features, sweep for hyperparameter optimization. MLflow — for on-premise deployment without external dependencies.

DVC (Data Version Control) — versioning of data and models on top of git. Data is stored in S3/GCS/Azure Blob, only metadata (hashes) in git. dvc repro reproduces the entire pipeline from raw data to metrics.

To ensure reproducibility of training, fix random seeds (torch.manual_seed, numpy.random.seed, random.seed) and record them in experiment metadata. Without this, debugging irregular results is painful. Log the dataset version (DVC hash) and git commit — then any experiment can be reproduced down to the byte.

Pipeline Orchestration: Kubeflow, Airflow, Prefect

A pipeline orchestrator becomes necessary when: A 100-line training script in cron is fine for simple tasks. But as soon as you have a multi-step pipeline (data loading → preprocessing → feature engineering → training → validation → deployment if quality above threshold), you need an orchestrator with retry logic, visualization, and alerts.

Kubeflow — Kubernetes-native orchestrator for ML (see Kubeflow). Each step is a Docker container. Supports parallel steps, conditional branches, artifacts between steps. Integrates with Katib (AutoML), KServe (serving), Feast (feature store).

Apache Airflow — more general DAG orchestrator. Wide ecosystem of operators (S3, Spark, DBT, Kubernetes). Easier to deploy if Airflow already exists in the company.

Prefect / Metaflow — less boilerplate. Prefect 2.x with @flow and @task decorators — quick start for small teams.

Typical training pipeline architecture on Kubeflow:

1. Data ingestion component — fetches data from S3/DB, validates schema via Great Expectations
2. Preprocessing component — transformations, normalization, train/val/test split
3. Training component — training on GPU, logging to MLflow
4. Evaluation component — metric calculation, comparison with baseline in Model Registry
5. Conditional deployment — deploy only if new model is better than current by >2% F1

Each component is a separate Docker image. Pipeline is versioned in git. Scheduled run (retraining once a week on new data) or manual.

Model Registry and Lifecycle Management

Model Registry is not just a checkpoint store. It is a centralized system that knows:

• Which model is currently in production (and with what metrics)
• History of all versions with training parameters
• Metadata: dataset, git commit, validation results
• Lifecycle stage: None → Staging → Production → Archived

MLflow Model Registry — standard. For enterprise — Vertex AI Model Registry (GCP), SageMaker Model Registry (AWS), Azure ML Model Registry.

Model promotion through stages: automatically move model to Staging after successful eval, then manual or automatic (during A/B test) promotion to Production. Rollback — switch to previous Production version in seconds.

Serving: From FastAPI to Triton Inference Server

Simple case. FastAPI + PyTorch/ONNX on one server — 80% of production ML deployments are exactly that. Sufficient for most tasks with load up to 100 req/s.

from fastapi import FastAPI
import onnxruntime as ort

app = FastAPI()
session = ort.InferenceSession("model.onnx", providers=["CUDAExecutionProvider"])

@app.post("/predict")
async def predict(request: PredictRequest):
    inputs = preprocess(request.text)
    outputs = session.run(None, {"input_ids": inputs})
    return {"label": postprocess(outputs)}

Triton Inference Server — production standard for high loads (500+ req/s). Dynamic batching, concurrent model execution, model ensemble. Supports TensorRT, ONNX, PyTorch TorchScript, TensorFlow SavedModel.

KServe — Kubernetes-native ML serving with autoscaling, canary deployments, A/B testing out of the box. Scale-to-zero for inactive models — savings on infrastructure up to 40% annually for a project with 10 models.

Monitoring: Data Drift, Model Drift, Infrastructure Metrics

Monitoring — what is usually done last and regretted first. Three levels.

Infrastructure monitoring. Latency (P50/P95/P99), throughput (req/s), error rate (4xx, 5xx), GPU/CPU utilization. Prometheus + Grafana — standard. Alert when P99 latency > threshold or error rate > 1%.

Data drift monitoring. Distribution of input data changes over time. Detect via PSI (Population Stability Index) for numerical features: PSI > 0.2 — strong drift. Chi-squared test for categorical, Kolmogorov-Smirnov test for continuous. Evidently AI — open source library with ready-made drift tests.

Model drift monitoring. If ground truth is delayed (e.g., we know conversion after a week) — monitor real metrics. If not — surrogate metrics: distribution of prediction scores, proportion of confident predictions.

Alerting. Three levels: INFO (minor drift, log it), WARNING (significant, notify team), CRITICAL (quality dropped below threshold — automatic switch to fallback model).

Why is data drift monitoring important?

Without it, you learn about model degradation only from user complaints or ringing SLA. A drift alert allows you to retrain the model in advance, before errors start causing losses. In one of our projects, PSI monitoring detected drift 2 days after a data source change — this saved the campaign.

Feature Store

Feature Store solves the training-serving skew problem. If preprocessing during training and inference is implemented in two different places — divergence is inevitable.

A Feature Store is needed when:

• Several models use the same features
• Features are computed from streaming data (real-time)
• Large team with different people on feature engineering and model training

Feast — open source Feature Store. Offline store (S3 + Parquet) for training, online store (Redis, DynamoDB) for low-latency inference. Feature definitions as code, materialization job syncs offline → online.

Tecton (commercial), Vertex AI Feature Store (GCP), SageMaker Feature Store (AWS) — managed options with less ops overhead.

CI/CD for ML

ML CI/CD is regular CI/CD plus specific ML steps.

ML-specific checks in CI:

• Reproducibility check: run training with a fixed seed, result must match
• Data validation: Great Expectations or Pandera on schema/distribution checks
• Model performance check: automatic eval on holdout, block merge if degradation > threshold
• Latency regression test: inference must meet SLA

GitOps for deployment. Merge to main → CI triggers training → eval → if passes → automatic deployment to Staging → smoke tests → manual promotion to Production or automatic upon successful canary.

Tools: GitHub Actions / GitLab CI for CI, ArgoCD for GitOps deployment on Kubernetes.

What's Included in MLOps Platform Development

We provide a full cycle of work, documentation, and team training.

Total time from audit to full MLOps platform: 3–5 months. Also possible phased launch: basic level (tracking + serving) in 4–6 weeks.

Cost is calculated individually based on data volume, number of models, and infrastructure requirements. Order an MLOps infrastructure audit — get a roadmap in 1–2 weeks. Contact us for a project assessment — we will send a preliminary estimate within 2 business days.

Note: warranty on architectural solutions — 12 months. We provide integration certificates with major cloud providers (AWS, GCP, Azure). During our work, we have not lost a single client after the first implementation — the experience of 50+ successful MLOps projects speaks for itself. Get a consultation on building an MLOps platform today.