Data Synthesis · Advanced · 20 min
Working With Fewer Than 100 Rows
Data SynthesisSmall data
The platform's home turf — the specific settings that work when data is scarce.
Steps
Follow along
Each step is something you can do in the platform right now.
01
Step 1
Upload your small dataset (tens of rows).
02
Step 2
Review column types extra carefully — with few rows, one mislabeled column dominates everything.
03
Step 3
Choose Matflow Statistical for the data synthesis mode; neural modes starve at this size.
04
Step 4
Generate a modest multiple (2–3×) of the original rows — not 10×.
05
Step 5
Evaluate and compare: if the quality report is weak, reduce rows or switch columns off.
06
Step 6
Train the predictor on the augmented set and read the extrapolation flags with extra attention.
Notes
Pro tips
What experienced users wish they knew the first time.
"Cope with" small data is not "perform as well as" large data — confidence intervals widen exactly when data is scarcest.
The demo datasets are a good sanity check: confirm the workflow works on a known-good file first.
Want the details? This tutorial pairs with the docs page models · synthesis models.
Read the docs →