clustersight 0.0.2

A package to analyze and interpret categorical clustered data

Clustering Explorer

The Clustering Explorer allows users to interactively analyze which factors in a dataset are most associated with clusters. Users can lasso points of interest in a 2D plot of the data, which is created using Principal Component Analysis (PCA) for dimensionality reduction. The tool provides three modes of analysis: 'table', 'histogram', and 'explainer'.

Usage

Create Lasso Tool

create_lasso(df, mode='table', label_col=None, exclude_cols=[], num_factors = 10, dtreeviz_plot=True)

The create_lasso function creates a lasso tool for data analysis. The parameters for this function are:

df: A Pandas DataFrame of the data to be analyzed mode: The mode of analysis. Can be 'table', 'histogram', or 'explainer' label_col: The column name to be used for color coding of the plot exclude_cols: A list of columns to exclude from the analysis num_factors: Number of factors to consider when mode is 'explainer' dtreeviz_plot: A boolean value to decide whether to plot decision tree using dtreeviz library

The mode parameter determines the type of analysis that will be performed:

'table': shows a table of the selected points 'histogram': shows an interactive histogram of each column's values among selected points compared with among all points 'explainer': predicts which factors lead to the clustered selection with a decision tree

The dtreeviz_plot parameter is used when mode is 'explainer'. If dtreeviz_plot is True, the decision tree is plotted using the dtreeviz library. Otherwise, the decision tree is plotted using sklearn, which is faster.

Dependencies

Python 3.6+ numpy pandas sklearn dtreeviz plotly ipywidgets itertools

Notes

The tool is designed for datasets that can fit in memory. For larger datasets, consider using a sampling method or dimensionality reduction techniques before using this tool.