bkamapantula / chart-recommender-gui

A rule-based chart recommendation service. Useful for teams to self-host or as an internal service.

Chart recommender

This project is adapted from suggest-chart by @tejesh0.

How it works

data file -> column type detection -> map against rules -> recommend charts

Rules

Rules are configured to work for a combination of numeric and categorical columns. num and cat refer to numeric and categorical columns, respectively.

Below charts are recommended for numeric and categorical columns:

Combination	Distribution of columns	Uniqueness	Charts
Numeric and Categorical columns	1 num, 1 cat	Unique values in categorical column	['lollipop', 'bar_plot', 'circular_bar_plot', 'treemap', 'circlepack']
Numeric and Categorical columns	1 num, 1 cat	Non-unique values in categorical column	['boxplot', 'violin_plot']
Numeric and Categorical columns	>1 num, 1 cat	Unique values in categorical column	['multi_line', 'parallel_plot', 'stacked_bar_plot', 'grouped_bar_plot']
Numeric and Categorical columns	>1 num, 1 cat	Non-unique values in categorical column	['grouped_scatterplot']
Numeric and Categorical columns	1 num, >1 cat	Unique values in categorical column	['multi_line', 'parallel_plot', 'stacked_bar_plot', 'grouped_bar_plot']. A variation of this is a nested view with these charts as possibilities: ['lollipop', 'bar_plot', 'circular_bar_plot', 'treemap', 'circlepack']
Numeric and Categorical columns	1 num, >1 cat	Non-unique values in categorical column	['box_plot', 'violin_plot']
Numeric and Categorical columns	1 num, >1 cat	Non-unique values in categorical column	Using Adjacency as a principle, these are the chart possibilities: ['network', 'sankey', 'chord', 'arc']

Below charts are recommended for numeric columns alone:

Combination	Distribution of columns	Uniqueness	Charts
Numeric	1 num	-	['histogram', 'density_plot']
Numeric	2 num	Unordered, few data points	['facet_box_plot', 'scatterplot']
Numeric	2 num	Unordered, many data points	['facet_violin_plot', 'facet_density_plot']
Numeric	2 num	Ordered	['line_chart', 'area_chart', 'connected_scatterplot']
Numeric	3 num	Unordered	['box_plot', 'violin_plot', 'bubble_plot']
Numeric	3 num	Ordered	['stacked_area_plot', 'line_graph']
Numeric	>3 num	Unordered	['box_plot', 'violin_plot', 'heatmap', 'correlogram']
Numeric	>3 num	Ordered	['stacked_area_plot', 'line_graph']

Below charts are recommended for numeric columns alone:

Combination	Distribution of columns	Type	Charts
Categorical	1 cat	-	['barplot', 'lollipop', 'donut', 'treemap', 'circlepack']
Categorical	>1 cat	Nested	['sunburst', 'treemap']
Categorical	>1 cat	Subgroup	['grouped_scatterplot', 'parallel_plot', 'stacked_bar_plot', 'grouped_bar_plot']
Categorical	>1 cat	Adjacency	['heatmap', 'network', 'sankey']

This work follows a subset of the excellent rule set defined by from Data to Viz.

Recommended charts

Here is the list of all possible chart recommendations:

• bar chart
• box plot
• bubble sort
• circlepack
• connected scatterplot
• correlogram
• density plot
• donut
• facet box plot
• facet density plot
• grouped scatterplot
• heatmap
• histogram
• line chart
• lollipop
• parallel coordinates
• network
• stacked bar chart
• stacked area chart
• sankey
• scatterplot
• sunburst
• treemap
• violin plot

Usage

Application setup

• Install Gramex 1.x
• Clone this repository
• From the repo folder, run gramex setup .
• From the repo folder, run gramex

Command-line interface (CLI)

Run recommend.py

python recommend.py

This uses data.xlsx file and recommends few charts as an output:

{'chart_list': ['lollipop', 'bar_plot', 'circular_bar_plot', 'treemap', 'circlepack']}

It runs initiate() function which uses two columns ID (categorical column with unique observations/values) and c1 (numeric column) to recommend charts.

As a service

import recommend

recommend.initiate('')

TODO

CLI tool

• Accept Google spreadsheet URL as input
• Accept local data file as input
• Accept a subset of columns

Standalone tool

• On running Gramex, provide an option to input a spreadsheet URL
• Recommend charts subsequently

Contributions

• Bhanu Kamapantula [e-mail]
• Tejesh [e-mail]
• Pragnya Reddy [e-mail]