Typical Workflow

Sequences → Distance Measure → Distance Matrix → Clustering Method → Choose k → Clusters

That chain is the classic sequence-clustering workflow. Sequenzo also supports visualization, group comparison, model-based analysis, multidomain work, event histories, feature extraction, and robustness checks. Most paths start from the same data object, then branch by research question.

Broader Workflow Map

Raw data
  ↓
Data cleaning and time-column setup
  ↓
SequenceData
  ↓
Explore: statistics, indicators, and visualization
  ↓
Choose the research path
  ├─ Compare whole trajectories → distances → clustering / CLARA / representatives
  │     (CLARA scales clustering to large data; representatives are typical observed sequences)
  ├─ Compare predefined groups → discrepancy analysis / group comparison / decomposition
  ├─ Model latent dynamics → HMM / MHMM / NHMM / MNHMM
  ├─ Work across domains → CAT / DAT / IDCD / scaling guide
  ├─ Study relational pairs → hierarchical and relational sequence analysis
  ├─ Mine event patterns → frequent subsequences / event dynamics
  ├─ Model event histories → sequence history / SAMM / spell survival
  ├─ Create downstream variables → representativeness / fuzzy memberships / feature selection
  └─ Check robustness → timing uncertainty and stability diagnostics

Choose the method family from the research question, not from the function list. Similar observed pathways? Start with distances. Predefined group differences? Start with discrepancy analysis or group comparison. Latent processes behind observed states? Start with Markov chain models.

Keep Method Choices Separate

Distance measures and clustering algorithms answer different questions.

Choice	What it decides	Where to continue
Distance measure	How two sequences are compared	Dissimilarity Measures
Clustering method	How similar sequences are grouped	Cluster Analysis Methods
Number of clusters	How coarse or detailed the typology should be	Cluster Quality Indicators

Think of the distance as the ruler and the clustering method as the grouping strategy. For example, CLARA is a scalable medoid-based clustering workflow. It still needs a distance measure, but it is not itself a distance measure.

For large datasets, use the CLARA guide after you have chosen the distance and clustering goal.

Choose a Next Path

If your question is...	Continue with...
How do I prepare my own data first?	Data Preprocessing Overview, then SequenceData
How do I compare whole trajectories?	Dissimilarity Measures, then get_distance_matrix()
How many clusters should I keep?	Cluster Quality Indicators, then ClusterQuality
What do the sequences look like before modeling?	Visualization Tools, Sequence Summary Statistics, and Sequence Characteristics Indicators
Do predefined groups explain trajectory differences?	Discrepancy Analysis and Group Comparison
Are there latent states or latent subgroups?	Markov Chain Models
Do several life domains unfold together?	Multidomain or Polyadic Sequence Analysis
Do repeated events form frequent patterns?	Event Sequences
Are transition histories or spell durations the focus?	Event History Analysis

Author: Yuqi Liang