Scaling Multidomain Workflows
Large multidomain projects need extra care because each person may have several aligned trajectories, such as work, family, and education. Sequenzo currently supports multidomain distances and combined-domain representations, while the CLARA shortcut is stable for single-domain SequenceData objects.
Use this page as a planning guide before starting a large multidomain analysis.
At a Glance
| Question | Guidance |
|---|---|
| Single-domain data too large for a full distance matrix? | Use clara(). |
| Multiple aligned domains? | Build CAT, DAT, or IDCD representations first. |
| Need clustering after IDCD? | Treat the IDCD result as the working sequence representation, then use the usual distance and clustering workflow. |
| Need CAT or DAT on very large data? | Start with uniqueness checks, sampling, and memory tests before computing full matrices. |
Recommended Order
- Check duplication first. Run
check_uniqueness_rate()for each domain. Repeated trajectories can make full distance computation much cheaper than the raw sample size suggests. - Choose a multidomain representation. Use IDCD for a combined-state sequence, CAT for aligned domain comparison, or DAT when each domain needs its own distance specification.
- Pilot on a subset. Estimate memory, runtime, and cluster stability before scaling to the full sample.
- Cluster only after the representation is clear. For IDCD-style workflows, continue with the standard distance matrix and clustering pages. For CAT or DAT, confirm that the full multidomain distance matrix is feasible.
- Report the constraint. In large multidomain studies, document sample size, number of time points, states per domain, uniqueness rates, and the scaling strategy.
Practical Choices
| Situation | Suggested path |
|---|---|
| One domain, many observations | clara() |
| Several domains with interpretable joint states | IDCD Sequence, then ordinary clustering |
| Several domains compared at the same time points | CAT Distance Matrix |
| Several domains with different distance settings | DAT Distance Matrix |
| Many rare combined states | Merge Sparse CombT Types before interpretation |
Important Limit
Do not import md_clara() or sequenzo.multidomain.clara in the current public package. Those names are not part of the released API. If you need a large multidomain analysis today, use the supported multidomain pages above and validate feasibility with a pilot run.
Related Pages
See Also
- Big Data Tools maps nearby scalability tools.
- Multidomain Overview explains the multidomain workflow.