# Latent Clusters Box

```{figure} ../../_static/images/tetrad-interface/box-by-box/latent-clusters-box.png
:name: tetrad-latent-clusters-box-screenshot
:alt: Latent Clusters Box in the Tetrad interface.

Latent Clusters Box in the Tetrad interface sidebar and main panel.
```

## Purpose

The **Latent Clusters** box is where you work with **clusterings and latent-cluster models** inferred from data.
These tools group observed variables (or sometimes cases) into clusters that are interpreted as being driven by
unobserved (latent) factors.

You use this box to:

- Run clustering or latent-cluster discovery procedures.
- Inspect which variables belong to which clusters.
- Create graphs or models that treat clusters as latent variables.

### Using latent clusters with Simulation

Cluster results by themselves only group variables; they do **not** yet specify a full causal structure
over the latent variables. To use latent clusters in **Simulation** mode, you typically:

1. Derive a latent structure (in the *Latent Structure* box) that introduces latent nodes and
   edges among them, using the clusters as indicators.
2. Use that latent structure model (possibly converted to a parametric model) as the source in
   the *Simulation* box.

See also:

- `Tetrad Interface → Detail Callouts → Latent Models and Simulation`


## Typical workflow

1. **Prepare data**
   - In the *Data* box, load and inspect the dataset you want to cluster.
   - Make sure variable types are appropriate for the clustering method you plan to use.

2. **Run a latent clustering method**
   - In the Latent Clusters box, choose:
     - The dataset.
     - A clustering or latent-variable discovery method (depending on what your Tetrad version exposes).
   - Configure method-specific options:
     - Number of clusters or range of clusters to consider (if required).
     - Any regularization or stopping criteria.
   - Click **Run** to perform clustering.

3. **Inspect clusters**
   - After the method finishes, select a latent clusters object from the list.
   - Use the main panel to view:
     - Which observed variables belong to which clusters.
     - Optional cluster-level summaries or statistics.
   - In some workflows, you can project these clusters into:
     - A graph where each cluster is represented as a latent node with edges to its member variables.
     - A parametric model (e.g., a multiple-indicator model).

4. **Use clusters in further modeling**
   - Export or convert the clustering result to:
     - A *Graph* with latent variables.
     - A *Parametric Model* or other downstream structure.
   - Use these derived structures in *Search*, *Simulation*, or *Estimator* as appropriate.

## Key controls

- **Toolbar**
  - **New / Configure** – set up a new latent clustering specification.
  - **Run** – execute the selected clustering method on the chosen dataset.
  - **Duplicate / Rename / Delete** – manage existing clustering results.
  - **Export** – save cluster assignments or derived structures to a file, if supported.

- **Latent clusters list**
  - Shows all latent-cluster objects in the project.
  - Each entry typically corresponds to:
    - A dataset,
    - A chosen method,
    - Specific parameter settings (e.g., number of clusters).

- **Main panel**
  - Displays details for the selected latent clusters object, such as:
    - A table mapping variables to clusters.
    - Cluster sizes and basic summaries.
    - Optional visualization or a derived graph view.

## Common patterns & tips

- Use latent clusters as a **preprocessing step** when you suspect that many observed variables are driven
  by a smaller number of latent factors.
- When evaluating cluster stability:
  - Run the same method with different random seeds or subsets of the data.
  - Compare assignments across runs using external tools or the *Compare* box (when applicable).
- Be cautious in interpretation:
  - Latent clusters are a modeling convenience and may or may not correspond to real-world constructs.
  - Always combine clustering results with domain knowledge.

## Related pages

- `Tetrad Interface → Overview` – high-level tour of the GUI.
- Other boxes that commonly interact with **Latent Clusters**:
  - *Data* (provides the dataset to be clustered).
  - *Graph* (may receive derived graphs with latent variables).
  - *Parametric Model* (can define SEMs or multiple-indicator models over discovered clusters).
  - *Latent Structure* (for more detailed latent-variable modeling).