Makeover

#30DayChartChallenge Day 3

Note

This post is my contribution to this year’s #30DayChartChallenge. Check out my Day 1 post to learn more.

Makeover

Makeover day has arrived, folks! Time to dust off those old relics from the ancient data catacombs and give them a fresh coat of paint. Today’s lucky candidate? A vintage scatter matrix plot taken from an introductory Data Science course I took many years ago.

The original scatter matrix plot, seen in the “Before” image, uses both histograms and scatter plots in a 4 panel chart to highlight the relationship between pairs of numerical variables. I would like to have a go at sprucing this chart up to make the insight the chart wants to share clearer.

Before

import pandas as pd
import matplotlib.pyplot as plt
drinks = pd.read_csv('https://raw.githubusercontent.com/justmarkham/DAT5/master/data/drinks.csv')

pd.plotting.scatter_matrix(drinks[['beer_servings', 'spirit_servings']])
plt.show()

After

from lets_plot import *
from lets_plot.bistro.joint import *
LetsPlot.setup_html()

drinks['continent'] = drinks['continent'].replace({
    'AS' : 'Asia',
    'EU' : 'Europe',
    'AF' : 'Africa',
    'NAm' : "North America",
    "SA" : "South America",
    "OC": "Oceania"
})

(
    joint_plot(
        data=drinks, 
        x='beer_servings', 
        y='spirit_servings', 
        color_by='continent',
        reg_line = False
        )
    + theme_minimal2()
    + labs(
        title = 'Distribution of average Beer vs Spirit servings across Continents',
        x = 'Beer Servings',
        y = 'Spirit Servings',
        caption = '#30DayChartChallenge #Day3 Makeover\nData: General Assembly DS Course\nMade by: www.ddanieltan.com'

    )
    + theme(
        legend_position='top',
        plot_caption=element_text(size=12, color='grey'),
        plot_title=element_text(size=18,),
    )
    + scale_color_brewer(palette='Dark2', guide=guide_legend(nrow=2))
    + ggsize(width=700,height=600)
)

Improvements made

1. Graphics are sharper courtesy of Lets-Plot’s more modern rendering engine
2. I like the use of a single panel instead of 4 for clarity
3. The line distributions in the margin of the chart mirror the relationship shown in the scatter plot, where we see most countries have low beer and spirit servings
4. The additional layer of mapping colour to continent adds more insights to the chart
5. Proper titles, axis labels and legends make a dramatic difference to readability

TIL

• I learnt the updated syntax to call Panda’s scatter matrix via its documentation
• I had the chance to explore Lets-Plot’s bistro plots documentation
• I discovered the guides layer for Lets-Plot, although for today’s plot, I decided to pass my guide param into the scale layer for brevity documentation