Making Plots With plotnine (aka ggplot)

Introduction

Python has a number of powerful plotting libraries to choose from. One of the oldest and most popular is matplotlib - it forms the foundation for many other Python plotting libraries. For this exercise we are going to use plotnine which is a Python implementation of the The Grammar of Graphics, inspired by the interface of the ggplot2 package from R. plotnine (and it's R cousin ggplot2) is a very nice way to create publication quality plots.

The Grammar of Graphics

Statistical graphics is a mapping from data to aesthetic attributes (colour, shape, size) of geometric objects (points, lines, bars)

Faceting can be used to generate the same plot for different subsets of the dataset

These are basic building blocks according to the grammar of graphics:

  • data The data + a set of aesthetic mappings that describing variables mapping
  • geom Geometric objects, represent what you actually see on the plot: points, lines, polygons, etc.
  • stats Statistical transformations, summarise data in many useful ways.
  • scale The scales map values in the data space to values in an aesthetic space
  • coord A coordinate system, describes how data coordinates are mapped to the plane of the graphic.
  • facet A faceting specification describes how to break up the data into subsets for plotting individual set

Let's explore these in detail.

First, install the pandas and plotnine packages to ensure they are available.

!pip install pandas plotnine

output


Requirement already satisfied: pandas in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (0.25.0)
Requirement already satisfied: plotnine in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (0.5.1)
Requirement already satisfied: numpy>=1.13.3 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from pandas) (1.17.0)
Requirement already satisfied: python-dateutil>=2.6.1 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from pandas) (2.8.0)
Requirement already satisfied: pytz>=2017.2 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from pandas) (2019.1)
Requirement already satisfied: descartes>=1.1.0 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from plotnine) (1.1.0)
Requirement already satisfied: scipy>=1.0.0 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from plotnine) (1.3.0)
Requirement already satisfied: patsy>=0.4.1 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from plotnine) (0.5.1)
Requirement already satisfied: matplotlib>=3.0.0 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from plotnine) (3.1.1)
Requirement already satisfied: statsmodels>=0.8.0 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from plotnine) (0.10.1)
Requirement already satisfied: mizani>=0.5.2 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from plotnine) (0.5.4)
Requirement already satisfied: six>=1.5 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from python-dateutil>=2.6.1->pandas) (1.12.0)
Requirement already satisfied: cycler>=0.10 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from matplotlib>=3.0.0->plotnine) (0.10.0)
Requirement already satisfied: kiwisolver>=1.0.1 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from matplotlib>=3.0.0->plotnine) (1.1.0)
Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from matplotlib>=3.0.0->plotnine) (2.4.1.1)
Requirement already satisfied: palettable in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from mizani>=0.5.2->plotnine) (3.2.0)
Requirement already satisfied: setuptools in /Users/perry/.virtualenvs/python-workshop-base-ufuVBSbV/lib/python3.6/site-packages (from kiwisolver>=1.0.1->matplotlib>=3.0.0->plotnine) (39.1.0)



# We run this to suppress various deprecation warnings from plotnine - keeps our notebook cleaner
import warnings
warnings.filterwarnings('ignore')

Plotting in ggplot style

Let's set up our working environment with necessary libraries and also load our csv file into data frame called survs_df,

import numpy as np
import pandas as pd
from plotnine import *

%matplotlib inline
survs_df = pd.read_csv('surveys.csv').dropna()

To produce a plot with the ggplot class from plotnine, we must provide three things:

  1. A data frame containing our data.
  2. How the columns of the data frame can be translated into positions, colors, sizes, and shapes of graphical elements ("aesthetics").
  3. The actual graphical elements to display ("geometric objects").

Introduction to plotting

ggplot(survs_df, aes(x='weight', y='hindfoot_length')) + geom_point()

png

output

Let's see if we can also include information about species and year.

ggplot(survs_df, aes(x='weight', y='hindfoot_length',
    size = 'year')) + geom_point()

png

output

Notice that we've dropped the x= and y= ? These are implied for the first and second argument of aes().

ggplot(survs_df, aes(x='weight', y='hindfoot_length', 
    size = 'year', color = 'species_id')) + geom_point()

png

output

We can do simple counting plot, to see how many observation (data points) we have for each year for example

ggplot(survs_df, aes(x='year')) + \
    geom_bar(stat = 'count')

png

output

Let's now also color by species to see how many observation we have per species in a given year

ggplot(survs_df, aes(x='year', fill = 'species_id')) + \
    geom_bar(stat = 'count')

png

output

Challenges

  1. Produce a plot comparing the number of observations for each species at each site. The plot should have site_id on the x axis, ideally as categorical data. (HINT: You can convert a column in a DataFrame df to the 'category' type using: df['some_col_name'] = df['some_col_name'].astype('category'))

  2. Create a boxplot of hindfoot_length across different species (species_id column) (HINT: There's a list of geoms available for plotnine in the docs - instead of geom_bar, which one should you use ?)

More geom types

ggplot(survs_df, aes(x='year', y='weight')) + \
    geom_boxplot()

png

output

Why are we not seeing mulitple boxplots, one for each year? This is because year variable is continuous in our data frame, but for this purpose we want it to be categorical.

survs_df['year_fact'] = survs_df['year'].astype("category")

ggplot(survs_df, aes(x='year_fact', y='weight')) + \
    geom_boxplot()

png

output

You'll notice the x-axis labels are overlapped. To flip them 90-degrees we can apply a theme so they look less cluttered. We will revisit themes later.

ggplot(survs_df, aes(x='year_fact', y='weight')) + \
    geom_boxplot() + \
    theme(axis_text_x = element_text(angle=90, hjust=1))

png

output

To save some typing, let's define this x-axis label rotating theme as a short variable name that we can reuse:

flip_xlabels = theme(axis_text_x = element_text(angle=90, hjust=1))

ggplot(survs_df, aes(x='year_fact', y='weight')) + \
    geom_violin() + \
    flip_xlabels

png

output

To save an image for later:

plt1 = ggplot(survs_df, aes(x='year_fact', y='weight')) + \
           geom_boxplot() + \
           xlab("Years") + \
           ylab("Weight log2(kg)") + \
           ggtitle("Boxplots, summary of species weight in each year")

ggsave(filename="plot1.png",
       plot=plt1,
       device='png',
       dpi=300,
       height=25,
       width=25)

Challenges

  1. Can you log2 transform weight and plot a "normalised" boxplot ? Hint: use np.log2() function and name new column weight_log.

  2. Does a log2 transform make this data visualisation better ?

Faceting

ggplot has a special technique called faceting that allows to split one plot into multiple plots based on a factor included in the dataset. We will use it to make one plot for a time series for each species.

ggplot(survs_df, aes(x='year_fact', y='weight')) + \
    geom_boxplot() + \
    facet_wrap(['sex']) + \
    flip_xlabels + \
    theme(axis_text_x = element_text(size=6))

png

output





ggplot(survs_df, aes(x='year_fact', y='weight')) + \
    geom_boxplot() + \
    theme(axis_text_x = element_text(size=4)) + \
    facet_wrap(['species_id']) + \
    flip_xlabels

png

output

The two faceted plots above are probably easier to interpret using the weight_log column we created - give it a try !

The "Layered Grammar of Graphics"

ggplot(data = <DATA>) + 
  <GEOM_FUNCTION>(
     mapping = aes(<MAPPINGS>),
     stat = <STAT>, 
     position = <POSITION>
  ) +
  <COORDINATE_FUNCTION> +
  <FACET_FUNCTION>

Theming

plotnine allows pre-defined 'themes' to be applied as aesthetics to the plot.

A list available theme you may want to experiment with is here: https://plotnine.readthedocs.io/en/stable/api.html#themes

ggplot(survs_df, aes(x='year_fact', y='weight')) + \
    geom_boxplot() + \
    theme_bw() + \
    flip_xlabels

png

output





ggplot(survs_df, aes(x='year_fact', y='weight_log')) + \
    geom_boxplot() + \
    facet_wrap(['species_id']) + \
    theme_xkcd() + \
    theme(axis_text_x = element_text(size=4, angle=90, hjust=1))

png

output

Let's try to bin years into decades, which could be crude but might gives simple images to look at.

bins = [(survs_df['year'] < 1980),
        (survs_df['year'] < 1990),
        (survs_df['year'] < 2000),
        (survs_df['year'] >= 2000)]

labels = ['70s', '80s', '90s', 'Z']

survs_df['year_bins'] = np.select(bins, labels)

plt2 = ggplot(survs_df, aes(x='year_bins', y='weight_log')) + \
           geom_boxplot()
plt2

png

output





plt2 = ggplot(survs_df, aes(x='year_bins', y='weight_log')) + \
           geom_boxplot() + \
           flip_xlabels + \
           facet_wrap(['species_id'])
plt2

png

output

This is a different way to look at your data

ggplot(survs_df, aes("year_fact", "weight")) + \
    stat_summary(fun_y = np.mean, fun_ymin=np.min, fun_ymax=np.max) + \
    theme(axis_text_x = element_text(angle=90, hjust=1))

ggplot(survs_df, aes("year_fact", "weight")) + \
    stat_summary(fun_y = np.median, fun_ymin=np.min, fun_ymax=np.max) + \
    theme(axis_text_x = element_text(angle=90, hjust=1))

ggplot(survs_df, aes("year_fact", "weight_log")) + \
    stat_summary(fun_y = np.mean, fun_ymin=np.min, fun_ymax=np.max) + \
    theme(axis_text_x = element_text(angle=90, hjust=1))

png

output