Weighted Random Sampler
Draw rows from a CSV or JSON dataset with probability proportional to a weight column — with or without replacement, reproducible by seed. Runs entirely in your browser, no upload, no sign-up.
About this tool
Weighted Random Sampler draws a random subset of rows from a dataset where some rows should appear more often than others. Each row carries a weight (a non-negative number in a column or JSON key), and the sampler picks rows with probability proportional to that weight — a row with weight 5 is five times as likely to be drawn as a row with weight 1. It works on CSV text or a JSON array of objects, and every draw is reproducible: the same seed and inputs always produce the same sample.
Use it to build weighted lotteries and giveaways, to over- or under-sample classes when preparing training data, to pick survey respondents proportional to a population weight, or to simulate a weighted draw for testing.
Worked example
Input CSV (weights in the weight column):
name,weight
Alice,5
Bob,1
Carol,3
Dan,1
Drawing 2 rows without replacement, seed=42, gives:
name,weight
Alice,5
Carol,3
Alice and Carol carry the most weight, so with this seed they win the draw. Change the seed for a different sample, or turn replacement on to let a heavy row be drawn more than once (then the sample size can exceed the number of rows).
With vs. without replacement
- Without replacement (default) — each row is drawn at most once, so the sample size can't exceed the number of rows that have a positive weight. Rows come back in their original order. Uses the Efraimidis–Spirakis weighted scheme.
- With replacement — each of the
ndraws is independent, so the same row can appear several times andnmay be larger than the dataset. Rows come back in draw order.
Limits & edge cases
- Weights must be non-negative numbers. A row with weight
0is never drawn; a non-numeric weight is an error that names the offending row. - If all weights are zero, there is nothing to sample and the tool errors.
- Without replacement, asking for more rows than have a positive weight is an error.
- CSV: the header row (when enabled) is preserved and used to resolve a named weight field; the chosen delimiter is used for both input and output. JSON: the input must be an array of objects and the output is a JSON array of the selected objects.
- Everything runs locally in your browser — the dataset never leaves your machine.
FAQ
How are the weights turned into probabilities?
Each row's chance of being drawn is its weight divided by the sum of all weights.
So in A=5, B=1, C=3, D=1 (total 10), A has a 50% chance per draw, C 30%, and B
and D 10% each. You don't need the weights to add up to 1 or 100 — any
non-negative numbers work, and they're normalized for you.
What's the difference between sampling with and without replacement?
Without replacement (the default) draws each row at most once, so you can't ask for more rows than exist, and the result keeps the input order. With replacement treats every draw independently, so the same row can be picked multiple times and you can draw more rows than the dataset contains — useful for bootstrap resampling or weighted dice-style draws.
Is the sample reproducible?
Yes. Sampling uses a small deterministic PRNG seeded by the seed field, so the same dataset, size, and seed always produce the exact same rows. Change the seed to get a different draw while keeping everything else fixed.
Can I use it with JSON instead of CSV?
Yes. Set Format to json and paste a JSON array of objects, then give the
weight field as the object key holding the weight (e.g. w). The result is a
JSON array containing the full selected objects. Weights may be JSON numbers or
numeric strings like "2.5".
What happens to rows with a zero or missing weight?
A weight of 0 means the row is never selected, but it stays in your dataset and
counts toward nothing. A missing weight field or a non-numeric value is treated as
an error that tells you which row is at fault, so you can fix the data rather than
get a silently skewed sample.
Developer & Automation Access
Run it from the terminal
Same engine as this page, headless — via the gizza CLI:
gizza tool weighted-random-sampler "name,weight
Alice,5
Bob,1
Carol,3
Dan,1" 'weight_field=weight'New to the CLI? Get gizza →
Open it by URL
Pre-fill and auto-run this tool with query parameters — the names match the API/CLI:
https://gizza.ai/tools/weighted-random-sampler/?data=name%2Cweight%0AAlice%2C5%0ABob%2C1%0ACarol%2C3%0ADan%2C1&format=csv&weight_field=weight&n=2&replacement=true&seed=42&header=true&delimiter=commaMachine-readable descriptor: tool.json — title + parameters JSON Schema for agents.