Summary of basic drawing method of the de facto standard Plotly Express of Python drawing library in the Reiwa era

at first

A library called Plotly Express (official site) that allows you to easily perform interactive operations based on javascript was released a while ago. I found. I felt the potential of a dimension clearly different from the static drawing so far, so I summarized how to draw a basic visualization method as a memorandum. For reference, I am comparing it with the pattern that I wrote a similar figure with matplotlib + seaborn.

The "basic visualization method" is based on the previously written "How to select and draw the correct visualization method in exploratory data analysis".

What is Plotly Express?

Plotly Express is a set of high-level APIs for plotly released in March 2019. The feature is that you can easily write interactive and complicated drawings.

Official site: https://plot.ly/python/plotly-express/ API reference: https://www.plotly.express/plotly_express/#plotly_express.line Github (part of plotly): https://github.com/plotly/plotly.py

Seaborn is a Wrapper for writing Matplotlib easily, but let's compare Plotly Express and Seaborn to see which is better. Roughly

• seaborn is a library for easy static drawing based on matplotlib
• Plotly Express is a javascript-based library for easy and dynamic drawing

It is a contrasting structure. (Reference) Summary of drawing libraries in 2018 stage organized by Anaconda

What is plotly

Overview

plotly is an open source and interactive data visualization tool. Besides Python, it can be used with R and javascript. From January 11, 2019, the version of plotly.py has been increased to 4. It is an MIT license. Github: https://github.com/plotly/plotly.py

Until plotly3, it seems that the created graph had a more commercial smell, such as being processed or published "online" if you were not careful. "Offline mode" is now the default in plotly4, expanding what you can do within the free frame and making it easier to use. Announcement of plotly 4.0.0: https://community.plot.ly/t/introducing-plotly-py-4-0-0/25639

(Reference) Plotly product group

** Free ** plotly: A visualization module based on javascript for interactive operation

** Paid (partial) ** Chart Studio: A service created by disconnecting plotly's online mode? Visualization system DASH: You can create dashboard apps in Python or R without the need for knowledge of javascript.

Module installation

Assuming that mac, jupyter lab is included Some settings are required to support jupyter lab

pip install plotly

# Avoid "JavaScript heap out of memory" errors during extension installation
# (OS X/Linux)
export NODE_OPTIONS=--max-old-space-size=4096

# Jupyter widgets extension
jupyter labextension install @jupyter-widgets/[email protected] --no-build

# FigureWidget support
jupyter labextension install [email protected] --no-build

# and jupyterlab renderer support
jupyter labextension install [email protected] --no-build

# JupyterLab chart editor support (optional)
jupyter labextension install [email protected] --no-build

# Build extensions (must be done to activate extensions since --no-build is used above)
jupyter lab build

# Unset NODE_OPTIONS environment variable
# (OS X/Linux)
unset NODE_OPTIONS

For details, refer to the install page on the official website.

Library import

import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
import seaborn as sns
import plotly.express as px
from plotly.subplots import make_subplots
import plotly.figure_factory as ff
import plotly.graph_objects as go
import plotly.io as pio
import japanize_matplotlib
import datetime
from matplotlib.dates import MonthLocator
from IPython.display import HTML

sns.set_style('darkgrid')
pio.templates.default = 'seaborn'
plt.rcParams['font.family'] = 'IPAexGothic'

%matplotlib inline
%config InlineBackend.figure_formats = {'png', 'retina'}

import matplotlib
import plotly
print(matplotlib.__version__) # 3.1.1
print(sns.__version__) # 0.9.0
print(plotly.__version__) # 4.2.1

#Use iris dataset
iris = sns.load_dataset('iris')
iris.head()

histogram

matplotlib+seaborn

fig, axes = plt.subplots(ncols=2, figsize=(16, 4))
sns.distplot(iris['sepal_length'], bins=np.arange(4,8,0.25), kde=False, label='all', ax=axes[0])
sns.distplot(iris.query('species=="setosa"')['sepal_length'], kde=False,
             bins=np.arange(4,8,0.25), label='setosa', hist_kws={'alpha':0.3}, ax=axes[1])
sns.distplot(iris.query('species=="versicolor"')['sepal_length'], kde=False,
             bins=np.arange(4,8,0.25), label='versicolor', hist_kws={'alpha':0.3}, ax=axes[1])
sns.distplot(iris.query('species=="virginica"')['sepal_length'], kde=False,
             bins=np.arange(4,8,0.25), label='virginica', hist_kws={'alpha':0.3}, ax=axes[1])
axes[0].set_title("Histogram of length(cm)")
axes[1].set_title("Histogram of the length of the sword for each variety(cm)")
axes[1].legend()
plt.show()

Plotly Express

#You can't explicitly specify the width of the bin, but changing the size of the nbins will give you a nice width.
#plotly express generates fig
#If you want to store the diagram in subplots, each fig.You can retrieve and store trace from data
#However, if marginal is specified, the position of marginal will collapse, so leave it for a while.
fig = px.histogram(iris, x='sepal_length', color='species', 
                           nbins=19, range_x=[4,8], width=600, height=350,
                           opacity=0.4, marginal='box')
#range when drawing histogram_If y is specified, the drawing position of the marginal box plot will be corrupted.
fig.update_layout(barmode='overlay')
fig.update_yaxes(range=[0,20],row=1, col=1)
#Save as html
# fig.write_html('histogram_with_boxplot.html', auto_open=False)
fig.show()

Click here for a working link: https://uchidehiroki.github.io/plotly_folders/Basic_Charts/histogram_with_boxplot_px.html

When you hover your mouse over it, various information will pop up. Marginal plot is easy to apply. You can see only that by double-clicking the variety you want to see from the item on the right.

plotly I will draw using the function of plotly body

fig = make_subplots(rows=1, cols=2, subplot_titles=('Sepal length(cm)', '品種毎のSepal length(cm)'))
fig.add_trace(go.Histogram(x=iris['sepal_length'], xbins=dict(start=4,end=8,size=0.25), hovertemplate="%{x}cm: %{y}Pieces", name="All varieties"), row=1, col=1)
for species in ['setosa', 'versicolor', 'virginica']:
    fig.add_trace(go.Histogram(x=iris.query(f'species=="{species}"')['sepal_length'],
                           xbins=dict(start=4,end=8,size=0.25), hovertemplate="%{x}cm: %{y}Pieces", name=species), row=1, col=2)
fig.update_layout(barmode='overlay', height=400, width=900)
fig.update_traces(opacity=0.3, row=1, col=2)
fig.update_xaxes(tickvals=np.arange(4,8,0.5), title_text='sepal_length')
fig.update_yaxes(title_text='frequency')
fig.write_html('../output/histogram_with_boxplot_plotly.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/histogram_with_boxplot_plotly.html

Box plot

matplotlib+seaborn

fig, ax = plt.subplots(figsize=(8,3))
sns.boxplot(x='sepal_length', y='species', data=iris, order=('setosa', 'versicolor', 'virginica'), ax=ax)
ax.set_title('Box plot of sepal length for each variety(cm)')
plt.show()

Plotly Express

fig = px.box(iris, y='species', x='sepal_length', color='species', orientation='h',
       category_orders={'species': ['setosa', 'versicolor', 'virginica']}, 
       title='Box plot of sepal length for each variety(cm)', width=600, height=400)
fig.update_layout(showlegend=False)
fig.write_html('../output/boxplot_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/boxplot_px.html For example, hovering the cursor over an outlier point will pop up information. It's not beautiful that each variety doesn't get on the horizontal line ...

plotly

fig = go.Figure()
for species in iris['species'].unique():
    fig.add_trace(go.Box(x=iris.query(f'species=="{species}"')['sepal_length'], name=species))
fig.update_layout(height=300, width=600, showlegend=False, title_text='Box plot of sepal length for each variety(cm)')
fig.update_xaxes(title_text='sepal_length')
fig.write_html('../output/boxplot_plotly.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/boxplot_plotly.html This is a boxplot on the line

bar graph

matplotlib + seaborn

fig, ax = plt.subplots(figsize=(8,3))
sns.barplot(x='sepal_length', y='species', order=['setosa', 'versicolor', 'virginica'], ci='sd', data=iris, ax=ax)
ax.set_title('Average length of sepals for each variety(cm)')
plt.show()

Plotly Express

#Unlike Seaborn, Plotly doesn't calculate the average length or standard deviation.
agg_iris = iris.groupby(['species'])[['sepal_length']].agg([np.mean, np.std, 'count']).reset_index()
agg_iris.columns = ['species', 'sepal_length', 'std', 'count']
fig = px.bar(agg_iris, x='sepal_length', y='species', color='species', category_orders={'species': ['setosa', 'versicolor', 'virginica']}, 
       error_x='std', orientation='h', hover_data=['count'], height=300, width=600, title='Average length of sepals for each variety(cm)')
fig.update_layout(showlegend=False)
fig.write_html('../output/barplot_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/barplot_px.html

(Reference) ggplot If you want to use facet to see

Useful when there are multiple comparison axes

tips = px.data.tips()
fig = px.histogram(tips, x="sex", y="tip", histfunc="avg", color="smoker", barmode="group",
             facet_row="time", facet_col="day", category_orders={"day": ["Thur", "Fri", "Sat", "Sun"],
                                                                "time": ["Lunch", "Dinner"]},
                  height=400, width=800)
fig.write_html('../output/boxplot_with_facet_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/boxplot_with_facet_px.html

Stacked bar graph

Preprocessing

iris['sepal_length_rank'] = pd.cut(iris['sepal_length'], [4,5,7,8], labels=['short', 'medium', 'long'])
agg_iris = iris.groupby(['sepal_length_rank','species']).size().reset_index()
agg_iris.columns = ['sepal_length_rank', 'species', 'count']
pivot_iris = agg_iris.pivot(index='sepal_length_rank', columns='species', values='count')

matplotlib+~~seaborn~~Pandas Plot

#stacked to seaborn=Since there is no function that corresponds to True, write it with Pandas plot
fig, ax = plt.subplots(figsize=(8,3))
pivot_iris.plot.barh(y=pivot_iris.columns, stacked=True, ax=ax)
ax.set_title('Frequency and variety breakdown of sepal length by rank')
ax.set_xlabel('frequency')
ax.set_ylabel('Sepal length rank')
plt.show()

Plotly Express

fig = px.bar(agg_iris, y='sepal_length_rank', x='count', color='species', orientation='h', barmode='relative',
      height=300, width=600, title='Frequency and variety breakdown of sepal length by rank')
fig.update_xaxes(title_text='frequency')
fig.update_yaxes(title_text='Sepal length rank')
fig.write_html('../output/stacked_barplot_px.html', auto_open=False)
fig

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/stacked_barplot_px.html You can select any element and stack it

Scatter plot

matplotlib + seaborn

#When you want to see the correlation between two variables
fig, ax = plt.subplots(figsize=(6,4))
sns.scatterplot(x='sepal_length', y='petal_length', hue='species', data=iris, ax=ax)
ax.legend(loc='lower right')
ax.set_title('Scatter plot of sepal length and petal length(cm)')
ax.set_xlabel('Sepal length(cm)')
ax.set_ylabel('Petal length(cm)')
plt.show()

#When you want to see the correlation of multiple variables
g = sns.pairplot(iris, hue='species')
plt.subplots_adjust(top=0.95)
g.fig.suptitle('Scatter plot matrix for each variety')
g.fig.set_size_inches(8,6)

Plotly Express

fig = px.scatter(px.data.iris(), x='sepal_length', y='petal_length', color='species', symbol='species', 
                 marginal_x='box', marginal_y='histogram', trendline='ols',
           hover_data=['species_id'], width=800, height=600, title='Scatter plot of sepal length and petal length(cm)')
#If range is not specified individually, not in scatter, it will be reflected in the marginal figure and the marginal figure will not be visible.
fig.update_xaxes(title_text='Sepal length(cm)', range=[4,8], row=1, col=1)
fig.update_yaxes(title_text='Petal length(cm)', range=[0.5,8], row=1, col=1)
fig.write_html('../output/scatterplot_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/scatterplot_px.html

fig = px.scatter_matrix(iris, dimensions=['sepal_length','sepal_width','petal_length','petal_width'],
                 color='species', size_max=1, title='Scatter plot matrix for each variety', width=800,height=600)
fig.write_html('../output/scattermatrix_px.html', auto_open=False)
fig

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/scattermatrix_px.html If you use the lasso tool in the upper right to enclose some data, only the data corresponding to that data will be highlighted in all figures. If you turn on "Toggle Spike Lines", it will give you auxiliary lines for x and y coordinates.

(Reference) Parallel coordinates

If there are too many variables, it is better to use parallel coordinates rather than looking at the scatterplot matrix

#Axis rearrangement can be operated interactively, this is dangerous
#Filtering is also possible for each axis, and the width shown can be adjusted by clicking the axis value.
#Rather than looking at each correlation with a scatterplot matrix, it seems better to look at these equilibrium coordinates because multiple axes can be considered.
fig = px.parallel_coordinates(px.data.iris(), 
                              color='species_id', 
                              dimensions=['sepal_length','sepal_width','petal_length','petal_width', 'species_id'],
                              color_continuous_scale=px.colors.diverging.Portland,
                             height=400, width=800)
fig.write_html('../output/parallel_coordinates_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/parallel_coordinates_px.html You can sort the axes If you select a value in the axis, it will be filtered only to the data that passes through that value.

#If the categorical variable is on the axis, you can do it like this
fig = px.parallel_categories(iris, dimensions=['species', 'sepal_length_rank'], color='sepal_length',
                             labels={'species': 'Ayame variety', 'sepal_length_rank': 'Sepal length grade'},
                            height=400, width=600)
fig.write_html('../output/parallel_categories_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/parallel_categories_px.html

Heat map

Preprocessing

flights = sns.load_dataset("flights")
#Cross tabulation
crossed_flights = pd.pivot_table(flights, values="passengers", index="month", columns="year", aggfunc=np.mean)  
# pd.crosstab(flights["month"], flights["year"], values=flights["passengers"], aggfunc=np.mean)But yes

matplotlib+seaborn

#Heat map
fig, ax = plt.subplots(figsize=(8,6))
sns.heatmap(crossed_flights,annot=True, cmap="Oranges", fmt='.5g', ax=ax)
ax.set_title('Passenger heat map');

Somehow the numbers go out of the frame ... It didn't happen in the past, but it's a mystery.

Plotly Express Since there was no corresponding function in Plotly Express, I will use the function of plotly itself plotly

fig = go.Figure()
fig.add_trace(go.Heatmap(z=crossed_flights, x=crossed_flights.columns, y=crossed_flights.index,
                        hovertemplate='%{x}-%{y}: %{z} passengers', colorscale='Oranges'))
fig.update_layout(height=400, width=600, title_text='Passenger heat map(With hover and without annotate)')
fig.write_html('../output/heatmap_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/heatmap_px.html

#The heatmap is easier to see if hover is turned off
fig = ff.create_annotated_heatmap(z=crossed_flights.values, x=list(crossed_flights.columns),
                                  y=list(crossed_flights.index), colorscale='Oranges',
                                  hoverinfo='none')
fig.update_layout(height=400, width=600, showlegend=False, title_text='Passenger heat map(Without hover With annotate)')
fig.write_html('../output/heatmap_with_annotate_px.html', auto_open=False)
fig

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/heatmap_with_annotate_px.html

Line graph

Preprocessing

#Prepare a datetime type column
datetimes = []
for i, j in enumerate(flights.itertuples()):
    datetimes.append(datetime.datetime.strptime(f'{j[1]}-{j[2]}', '%Y-%B'))
flights['datetime'] = datetimes

matplotlib+seaborn

fig, ax = plt.subplots(figsize=(12, 4))
sns.lineplot(x='datetime', y='passengers', data=flights, ax=ax)
ax.xaxis.set_major_locator(MonthLocator(interval=6))
ax.tick_params(labelrotation=45)
ax.set_xlim(datetime.datetime(1949,1,1,0,0,0),datetime.datetime(1961,1,1,0,0,0))
ax.set_title('Line graph of changes in the number of passengers(Monthly)')
plt.show()

Plotly Express

fig = px.line(flights, x='datetime', y='passengers', 
              height=400, width=800, title='Line graph of changes in the number of passengers(Monthly)')
fig.update_layout(xaxis_range=['1949-01-01', '1961-01-01'], # datetime.May be specified by datetime
                 xaxis_rangeslider_visible=True)
fig.update_xaxes(tickformat='%Y-%m', tickangle=45)
fig.write_html('../output/lineplot_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/lineplot_px.html You can check the line graph in the range specified by the slider below. When comparing multiple line charts, you can later select only the items you want to compare

Other interesting graphs

I was looking at https://plot.ly/python/plotly-express/ and arranged some interesting figures.

Polar Coordinates In the bar graph, you could only compare the length of univariate, but it seems that you can use it when you want to compare scores on multiple axes such as test results.

melted_iris = iris.melt(id_vars=['species'], value_vars=['sepal_length', 'sepal_width', 'petal_length', 'petal_width'],
                       var_name='variable', value_name='value')
melted_iris = melted_iris.groupby(['species', 'variable']).mean().reset_index()
fig = px.line_polar(melted_iris, r='value', theta='variable', color='species', line_close=True,
                   height=500, width=500, title='Length and width of sepals and petals for each variety of Ayame')
fig.write_html('../output/line_polar_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/line_polar_px.html Especially the shape of the setosa flower seems to be different

distribution

fig = make_subplots(rows=1, cols=2, subplot_titles=('Sepal length(cm)', '品種毎のSepal length(cm)'))
fig.add_trace(px.density_contour(iris, x="sepal_width", y="sepal_length").data[0],row=1, col=1)
fig2 = px.density_contour(iris, x="sepal_width", y="sepal_length", color='species')
[fig.add_trace(fig2.data[i], row=1, col=2) for i in range(len(fig2.data))] #Store all fig2 traces in fig
fig.write_html('../output/densityplot_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/densityplot_px.html Cool You can see with the mouse how high each contour line represents

animation

gapminder = px.data.gapminder()
fig = px.scatter(gapminder, x="gdpPercap", y="lifeExp", animation_frame="year", animation_group="country",
           size="pop", color="continent", hover_name="country", facet_col="continent",
           log_x=True, size_max=45, range_x=[100,100000], range_y=[25,90])
fig.write_html('../output/scatter_animation_px.html', auto_open=False)
fig.show()

https://uchidehiroki.github.io/plotly_folders/Basic_Charts/scatter_animation_px.html I think this is interesting when you click the URL and actually move it

General comment

I will summarize the results of comparing Plotly Express and seaborn (matplotlib) with my own judgment and prejudice. ◎: Recommended for use ○: No particular problem △: Dissatisfied ×: No implementation

Plotly Express's overwhelming victory There were many other drawings such as radar charts, density distributions, and animations that would be interesting if used well. Besides, if you can master Slider, it will be very cool. https://plot.ly/python/sliders/ https://uchidehiroki.github.io/plotly_folders/Basic_Charts/histogram_with_slider.html

(Digression 1) Points to note when using Plotly Express

• ** Data is retained in the document **
• Be careful when publishing as it is embedded in the script tag in json format
• It becomes heavy when the amount of data is large
• Browser + machine dependent
• For processing on the client side

From kazutan.R

It is more troublesome than the image file, but it seems easier to refer to it after placing the html file locally or hosting it online. It's better to host online, especially when sharing with multiple people

(Digression 2) How to refer to the html file later

Compared to png and jpeg image files, it takes a little trick to display html files. ** When displaying offline ** You can draw by specifying the path of the html file

** When displaying online ** plotly style: use chart_studio You can draw by hosting the html file on plotly's server and referencing the published URL However, free accounts do not allow you to issue secret links or host them in a private environment (rather, plotly makes money there). The public host of the html file also supports only up to 100 in total.

So use github pages. Place the html file in the github repository, select master branch in Settings → GitHub Pages → Source, and press the Save button. あとは、username.github.io/path/to/html/filesにアクセスすればOKです It's easy to manage html files because it's github. Official site: https://pages.github.com/ Reference: https://www.tam-tam.co.jp/tipsnote/html_css/post11245.html

It seems that Github pages themselves can be published even in a private repository. If you analyze with data that you do not want to publish to the public, it seems good to save the drawing html file in a private repository and share only the URL.

(Digression 3) HTML file reference in jupyter lab

#You can refer to the html file and embed it
HTML(filename='../output/histogram_with_boxplot_px.html')
#You can also refer to the static html you put online
HTML(url='https://uchidehiroki.github.io/plotly_folders/Basic_Charts/barplot_px.html')

At the end

I used to be a seaborn comedian, but I don't need to use it much anymore. I'm a little lonely, but I'm going to keep up with the times. I hope that the number of Plotly Express users will increase. It's been a long time, but thank you for your relationship.