Cohort analysis, from the first filter to the last comparison.
Many cohorts, many attributes, one tree.
CoralDuck is a cohort analysis tool that runs entirely in your browser. Load your dataset, define and curate cohorts, and compare them.
All your data is stored and processed locally, in your browser.
What you can do with it
Create cohorts by filtering and splitting, then characterize them: see their distributions, compare them statistically, and inspect them down to a single row.
Bring your own data
CoralDuck ships no data of its own. Point it at a CSV, TSV, Parquet or JSON file on your disk, or at a URL. A CSV or JSON file can come gzipped.
A tree that remembers
Every cohort keeps the steps that made it. The tree shows how your cohorts evolved, so you can always read where one came from — and go back and change it.
Charts that fit the data
Bar charts for categories, density plots and box plots for numbers, Kaplan-Meier curves for survival. All cohorts share the same bins, so they stay comparable.
Many cohorts, many attributes
Work on several cohorts at once, across several attributes. One action applies to every cohort you selected, and each gets its own result.
Prevalence at a glance
See how large a cohort is against a reference you choose, so a small group stays readable next to the population it came from.
Statistical tests built in
Compare cohorts pair by pair across attributes and read the p-values right in the comparison, without exporting anything.
How it works
Load your table once, then loop: create cohorts by filtering and splitting, characterize them, and let what you learn shape the next cut.
-
Load a table
The whole file becomes your first cohort, and everything descends from it.
-
Select and refine
Pick bars in a bar chart, a range in a density or box plot, a point on a survival curve. Filter keeps your selection; split makes one cohort per selection.
-
Analyze several at once
Look at multiple cohorts across multiple attributes side by side — down to the single rows behind them.
-
Estimate prevalence, then compare
Read how large each cohort is against a reference you define, then test every pair on every attribute.
The guide takes each step further
The tree keeps every step, so a cohort is always the record of how it was made — and reopening the project on another machine gets you the same tree.
Open CoralDuck