Menu
Discover natural groupings in data using clustering algorithms with evaluation and visualization. Use when segmenting data or finding patterns.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Analyze an A/B test experiment and provide a clear ship/iterate/no-ship recommendation. Use when evaluating experiment results.
Run correlation analysis on connected data, identify strong relationships, and suggest regression models. Use when exploring relationships between variables.
Run an end-to-end data analysis pipeline โ connect, profile, analyze, and report. Use when performing a complete analysis workflow.
Connect to a data source, profile its schema and quality, and recommend appropriate analyses. Use when starting work with a new dataset.
Analyze a time series and generate forecasts with confidence intervals. Use when predicting future values from historical data.
Explore a graph or network dataset โ structure, statistics, paths, and visualization. Use when working with DOT, GML, GraphML, or Mermaid files.
| name | cluster-analysis |
| description | Discover natural groupings in data using clustering algorithms with evaluation and visualization. Use when segmenting data or finding patterns. |
| allowed-tools | mcp__localdata__describe_database mcp__localdata__execute_query mcp__localdata__analyze_clusters mcp__localdata__reduce_dimensions |
| argument-hint | <database-name> |
Find natural groupings in data using multiple clustering approaches, evaluate quality, and interpret results.
Explore features. Call describe_database with the database name from $ARGUMENTS. Identify numeric columns suitable for clustering. Note any categorical columns that could provide context for interpreting clusters later.
Extract and review data. Call execute_query to select the numeric feature columns. Check for nulls and extreme outliers in the sample. Note the number of observations and features.
Run K-Means clustering. Call analyze_clusters with the database name, feature columns, and algorithm set to "kmeans". Start with k=3 unless domain knowledge suggests otherwise. Review the silhouette score and cluster sizes.
Try alternative k values. If the silhouette score is below 0.5, re-run analyze_clusters with k=2, k=4, and k=5. Compare silhouette scores to find the optimal number of clusters.
Run DBSCAN for comparison. Call analyze_clusters with algorithm set to "dbscan". This density-based approach does not require specifying k and can find irregularly shaped clusters. Compare the number of clusters found and the noise point percentage.
Evaluate and compare. Assess both approaches:
Reduce dimensions for visualization. Call reduce_dimensions with the database name and feature columns, using PCA with 2 components. This provides a 2D representation of the clusters for interpretation.
Interpret clusters. For the best clustering result, describe each cluster by its feature averages. Give each cluster a descriptive label based on its defining characteristics. Note which features most differentiate the clusters.
Present results. Provide:
Recommend next steps. Suggest using cluster labels as a feature in /localdata-mcp:regression, running /localdata-mcp:analyze-correlations within each cluster, or applying the clustering to new data for segmentation.
