Titanic dataset details
Supplemental source, numeric teaching subset and survival label details for dataset.titanic().
For loading and examples, start with Well known datasets. This page documents the exact source and numeric representation used by dataset.load("titanic") and dataset.titanic().
Source and attribution
Harrell Jr., F. E., and Cason, T. Titanic passenger data, titanic3. Vanderbilt Biostatistics datasets and the original CSV, retrieved October 13, 2026. Data obtained from http://hbiostat.org/data courtesy of the Vanderbilt University Department of Biostatistics. Source references include Eaton and Haas (1994), Titanic: Triumph and Tragedy, and Encyclopedia Titanica, as described in the source notes.
Vanderbilt grants permission to use its datasets and requests attribution. This is recorded as Vanderbilt-Data-Permission, a custom permission, not a Creative Commons license. See the retained permission notice.
| Property | Value |
|---|---|
| Variant | Vanderbilt titanic3, all 1,309 passenger records |
| Task | Binary survival classification |
| Runtime predictors | 4 |
| Classes | not_survived = 0 (809), survived = 1 (500) |
Upstream CSV SHA-256 | db6df9666818c69a753cd85d743e01502c8518b00579b6aada0d4fc5a66ccb9d |
| Runtime input | titanic.data, headerless predictors followed by target |
Runtime input SHA-256 | f2cc2aea1482f73ae0fa3a9fb99807fd816beb23ebd557cc407a2766e8f0bffd |
| Version | vanderbilt-titanic3-numeric-sha256-f2cc2aea1482 |
Numeric teaching representation
| Index | Feature | Encoding / unit |
|---|---|---|
| 0 | pclass | Passenger class, 1, 2, or 3; ordinal code |
| 1 | sex | Source categories encoded female = 0, male = 1 |
| 2 | sibsp | Number of siblings or spouses aboard, count |
| 3 | parch | Number of parents or children aboard, count |
All predictors are returned as Float64. Source row order and every passenger record are preserved. No observations are removed, no missing values are filled, and no statistics are fitted during loading. The survival target is excluded from features.
This is an explicitly selected numeric subset of the full source schema, not the Kaggle 891 row training file. Age is omitted because 263 records lack it; fare is omitted because one record lacks it. String and nullable columns await broader dataset support. Boat and recovered body fields are excluded because they reveal outcomes. The unmodified CSV remains available for inspection.
The loader does not infer families or attach group keys. Attach groups explicitly if the experiment requires related passengers to stay in the same partition. No official train/test assignment is imposed. Small historical teaching data do not establish general deployment performance.