Table of Contents

plotly

plotly can either render interactive graphs inside a jupyter notebook or save plots as html and open them inside a browser.

we import two plot functions from plotly.offline, plot for offline plotting and iplot for online plotting

Imports

# plotly
import plotly.plotly as py
import plotly.graph_objs as go
import plotly.tools as tls 
from plotly.offline import download_plotlyjs, init_notebook_mode, plot, iplot, iplot_mpl

# matplotlib, seaborn
import seaborn as sns
from matplotlib import cm
from matplotlib import pyplot as plt

Minimal Example

init_notebook_mode(connected = True)

trace = { 'x':[1,2], 'y':[1,2] }

data = [trace]

fig = go.Figure( data = data, layout = {} )

iplot(fig)

The code above would be sufficient to generate show an interactive java script graph in a jupyter notebook. However java script get lost when notebooks are converted first to .md and then to .html therefore we use an iframe to embedd the graph instead.

from IPython.display import HTML
HTML('<iframe width="900" height="800" frameborder="0" scrolling="no" src="//plot.ly/~datistics/1.embed"></iframe>')

Scatterplot

as in matplotlib we have to add each group of elements. This time we use a for loop to iterate ofer the unique species and then store the traces in the data list. For the colors we get the Brewer palettes from matplotlib importing cm (color maps). plotly excepts either named colors (‘red’, ‘green’, etc), HEX strings (‘#FF0000’) or rgba/rgb strings ‘rgba(255,0,0,1)’ as colors. Note that in python 3 the old % format operator for strings is deprecated.

df = sns.load_dataset('iris')

species = list( df.species.unique() )

colors_rgba = list( cm.Dark2( range(0,len(species),1) ) )
colors_str = [ 'rgba({},{},{},{})'.format(r,g,b,a) for r,g,b,a in colors_rgba ]

data = []

for i, spec  in enumerate( species, 0 ) :
    
    df_spec = df.loc[ df['species'] == spec, : ]
    
    trace = go.Scatter( x = df_spec.petal_length
                      , y = df_spec.petal_width
                      , mode = 'markers'
                      , name = spec
                      , marker = dict( color = colors_str[i] )
                      )    
    data.append(trace)
    

fig = go.Figure( data = data, layout = {} )

iplot(fig)

HTML('<iframe width="900" height="800" frameborder="0" scrolling="no" src="//plot.ly/~datistics/3.embed"></iframe>')

Convert matplotlib object to plotly object

In R we can convert most ggplot2 plots easily to plotly plots. We can do something similar in python.

there are a few tweaks to consider which we cannot cover all here, but check out this (tutorial)[http://nbviewer.jupyter.org/github/plotly/python-user-guide/blob/master/s6_matplotlylib/s6_matplotlylib.ipynb]

Minimal Scatterplot

fig = plt.figure()

plt.plot(list( df.petal_length ), list(df.petal_width), 'o' )

init_notebook_mode()

py_fig = tls.mpl_to_plotly(fig)

iplot_mpl(fig)

HTML('<iframe width="400" height="300" frameborder="0" scrolling="no" src="//plot.ly/~datistics/5.embed"></iframe>')

Scatterplot iteratively constructed

fig = plt.figure()

for i, spec  in enumerate( species, 0 ) :
    
    df_spec = df.loc[ df['species'] == spec, : ]
    
    plt.plot( list( df_spec.petal_length ), list(df_spec.petal_width), 'o' )


init_notebook_mode()

py_fig = tls.mpl_to_plotly(fig)

iplot_mpl(fig)

HTML('<iframe width="400" height="300" frameborder="0" scrolling="no" src="//plot.ly/~datistics/7.embed"></iframe>')

Scatterplot via pandas

Here we also add a legend, but the legend does not translate well to the plotly object, which we would have to edit manually to correct it.

# old school

fig, ax = plt.subplots()

df.loc[ df['species'] == 'setosa', : ].plot.scatter('petal_length', 'petal_width', label = 'setosa', color = 'blue', ax = ax)

# functional indexing
df.query('species == "versicolor"') \
    .plot.scatter( 'petal_length', 'petal_width'
                  , label = 'versicolor'
                  , color = 'orange' 
                  , ax = ax )
    
df.query('species == "virginica"') \
    .plot.scatter( 'petal_length', 'petal_width'
                  , label = 'virginica'
                  , color = 'green' 
                  , ax = ax ) 
    
init_notebook_mode()

py_fig = tls.mpl_to_plotly(fig)

# remove matplotlib default styling
iplot_mpl(fig, strip_style= True)

# keep matplotlib default styling
iplot_mpl(fig, strip_style= False)

C:\anaconda3\lib\site-packages\plotly\matplotlylib\renderer.py:445: UserWarning:

Dang! That path collection is out of this world. I totally don't know what to do with it yet! Plotly can only import path collections linked to 'data' coordinates

C:\anaconda3\lib\site-packages\plotly\matplotlylib\renderer.py:481: UserWarning:

I found a path object that I don't think is part of a bar chart. Ignoring.

HTML('<iframe width="400" height="300" frameborder="0" scrolling="no" src="//plot.ly/~datistics/9.embed"></iframe>')

HTML('<iframe width="400" height="300" frameborder="0" scrolling="no" src="//plot.ly/~datistics/9.embed"></iframe>')

Scatterplot via seaborn

Nice, but we loose the legend

sns.lmplot(x = 'petal_length', y = 'petal_width', data = df
           , hue = 'species'
           , fit_reg = False)

init_notebook_mode()

# we need to get the figure like this for some reason
fig = plt.gcf()

py_fig = tls.mpl_to_plotly(fig)

iplot_mpl(fig, strip_style= False)

HTML('<iframe width="400" height="400" frameborder="0" scrolling="no" src="//plot.ly/~datistics/13.embed"></iframe>')

Boxplot via seaborn from wide format

we loose the boxes

sns.boxplot(data=df)

init_notebook_mode()

# we need to get the figure like this for some reason
fig = plt.gcf()

py_fig = tls.mpl_to_plotly(fig)

iplot_mpl(fig, strip_style= False)

HTML('<iframe width="400" height="800" frameborder="0" scrolling="no" src="//plot.ly/~datistics/15.embed"></iframe>')

Boxplot via seaborn from long format

we still loose the boxes

df_melt = df.melt(value_vars=['sepal_length', 'sepal_width', 'petal_length', 'petal_width']
                 , id_vars = 'species')


sns.boxplot('variable', 'value', data = df_melt, hue = 'species')

init_notebook_mode()

# we need to get the figure like this for some reason
fig = plt.gcf()

py_fig = tls.mpl_to_plotly(fig)

iplot_mpl(fig, strip_style= False)

C:\anaconda3\lib\site-packages\plotly\matplotlylib\renderer.py:481: UserWarning:

I found a path object that I don't think is part of a bar chart. Ignoring.

HTML('<iframe width="400" height="300" frameborder="0" scrolling="no" src="//plot.ly/~datistics/17.embed"></iframe>')

Violin plots via seaborn

Does not work

# sns.violinplot('variable', 'value', data = df_melt
#               , hue = 'species'
#               , inner = None ## removes inner boxes
#               , zorder = 1
#              )
#
# init_notebook_mode()
#
# # we need to get the figure like this for some reason
# fig = plt.gcf()
#
# py_fig = tls.mpl_to_plotly(fig)
#
# iplot_mpl(fig, strip_style= False)

Factor Plot via seaborn

ax = sns.factorplot('variable', 'value', data = df_melt
              , hue = 'species'
              , col = 'species'
              , kind = 'box' )

ax.set_xticklabels(rotation = -45)

init_notebook_mode()

# we need to get the figure like this for some reason
fig = plt.gcf()

py_fig = tls.mpl_to_plotly(fig)

iplot_mpl(fig, strip_style= False)

C:\anaconda3\lib\site-packages\plotly\matplotlylib\renderer.py:516: UserWarning:

Looks like the annotation(s) you are trying 
to draw lies/lay outside the given figure size.

Therefore, the resulting Plotly figure may not be 
large enough to view the full text. To adjust 
the size of the figure, use the 'width' and 
'height' keys in the Layout object. Alternatively,
use the Margin object to adjust the figure's margins.

C:\anaconda3\lib\site-packages\plotly\matplotlylib\renderer.py:481: UserWarning:

I found a path object that I don't think is part of a bar chart. Ignoring.

HTML('<iframe width="900" height="300" frameborder="0" scrolling="no" src="//plot.ly/~datistics/19.embed"></iframe>')

Summary

plotly plots look great if we use the original synthax. Converting matplotlib objects to plotly format is not worth it. They plotly and the matplotlib synthax have in common that they are quite cumbersome and that we need to use loops or very long repetitive code to pupulate the graphs. seaborn tackles this by reducing looping and providing excellent default settings. However we will occasionally encounter glitches that we need to tackle by iteratively reconfiguring attributes of plot elements in matplotlib.