Worlasidzam/KENTE

An interactive dataset structure explorer built with Streamlit # KENTE **Revealing the hidden structure of data.** KENTE is an interactive exploratory data analysis tool that transforms a dataset into a visual map of relationships, clusters, and dependencies between variables. Instead of manually running dozens of exploratory steps, users upload a CSV and receive an immediate structural explanation of their data. The name comes from Kente cloth, a traditional Ghanaian textile where individual threads weave together to form complex, meaningful patterns — just as KENTE reveals how individual variables connect to form the structure of a dataset. ## The Problem When working with a new dataset, analysts face a common set of questions: Which variables are truly related? Which relationships are direct and which are mediated through other variables? Are there hidden subgroups in the data? Are some variables redundant? Answering these questions typically requires hours of manual exploratory analysis across multiple tools and techniques. Most of this work follows the same pattern every time, yet there are few tools that unify these steps into a single coherent view. ## What KENTE Does KENTE combines multiple analytical lenses into one interactive workflow: **Correlation analysis** identifies which variables move together and flags potentially redundant pairs. **Conditional dependence estimation** uses the graphical lasso to recover the precision matrix, revealing direct relationships that persist after accounting for all other variables — separating genuine structure from spurious associations. **PCA and clustering** expose structure among observations, identifying subgroups and outliers in reduced space. Beyond standard EDA, KENTE introduces several features designed to surface findings that traditional tools miss: - **Structural Contradictions** — automatically detects *phantom links* (variables that appear correlated but have no direct dependency) and *hidden links* (variables with weak correlation but …