liveBook · Manning

Synthetic data

As explained in Vibe Engineering

Synthetic data is data generated algorithmically rather than directly collected from the underlying real-world process.

Synthetic data is generated algorithmically rather than collected from the underlying real-world process. Its cleanliness and uniform structure can speed up prototypes, but synthetic distributions may not represent real markets or preserve useful ground truth. section 4.1

Use synthetic data selectively:

  • Choose it for mock data, demonstrations, algorithm education, and testing machine-learning workflows. section 4.1
  • Treat it cautiously for exploratory analysis, market studies, demand forecasting, or other tasks requiring real-world distributions. section 4.1
  • Check whether generated fields have meaningful semantics, realistic distributions, and ground truth before treating outputs as actionable. section 4.1

Medicine datasets

A synthetic medicine dataset with about 50,000 entries may be cleaner and easier for controlled model demonstrations. A scraped Indian medicine dataset with about 253,973 records is harder to clean but better suited to real pricing, manufacturer, composition, and market analysis. The choice follows the question: prototype the workflow with synthetic data; study the pharmaceutical market with real data. section 4.1

Use Case
Dataset A (Indian)
Dataset B (Synthetic)
Exploratory Data Analysis (EDA)
✅ Strong — real distributions, price trends
⚠️ Limited — synthetic distributions
Supervised ML for Prediction
⚠️ Limited (no target label by default)
⚠️ Same — likely no target labels
Clustering / Unsupervised learning
✅ Yes — cluster by manufacturer, composition
⚠️ Yes but less real-world meaning
NLP/Categorization
⚠️ Yes — text fields (names/composition)
⚠️ Yes (names) but synthetic semantics
Recommender systems
⚠️ Possible but needs augmentation
❌ Limited
Benchmarking models / DEMO
⚠️ Possible but raw
✅ Ideal due to cleaner, controlled data
Aspect
Indian Medicine Dataset (A)
Kaggle Medicine Dataset (B)
Source
Scraped real data
Synthetic generated data
Size
~250 K records
~50 K records
Real-world
✅ Yes
❌ No
Cleanliness
⚠️ No (requires cleaning)
✅ Likely
Usability for Research
High
Moderate
ML Power
Good for real analysis
Good for ML practice

Compare synthetic and scraped data across exploratory analysis, prediction, clustering, NLP, recommender systems, benchmarking, size, cleanliness, and research value. section 4.1

How Vibe Engineering covers it

How deep it goes
Compares generated and real datasets through cleanliness, semantics, distributions, ground truth, and analytical suitability, without explaining generation mechanics.
Examples
Uses a 50,000-record synthetic medicine dataset contrasted with a roughly 253,973-record scraped Indian medicine dataset.
What it assumes
Assumes a business question and basic programming knowledge, while making Python optional for creating an analytics notebook.
What you can do after
Evaluate dataset suitability for prototyping, demonstrations, workflow testing, or domain-grounded analysis.
How it connects
Connects synthetic data to exploratory data analysis, supervised machine learning, clustering, NLP, recommender systems, and benchmarking.

Also explained in

Compare how 24 books explain synthetic data →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Vibe Engineering. The section links point to the parts of the book it draws on.