Notebook

imports and functions¶

This code borrows heavily from bokeh after finding out that they have implemented the function export_png to, well, export png files from their web-based plots. This has been requested many times of plotly with no solution so I tried applying what bokeh does... and it seems to work!

requests/mentions via plotly github: 351, 352, 483, 564, 596, 856, 880
this question at the plotly community forums
this and this on StackOverflow

In [34]:

import io
import pandas as pd
import plotly
import plotly.graph_objs as go
import selenium.webdriver
import shutil
import time

from os.path import devnull
from PIL import Image
from subprocess import Popen, PIPE

        
### from bokeh/util, slightly modified to avoid using bokeh's settings.py
### - https://github.com/bokeh/bokeh/blob/master/bokeh/util/dependencies.py
def detect_phantomjs():
    '''Detect if PhantomJS is avaiable in PATH.'''
    try:
        phantomjs_path = shutil.which('phantomjs')
    # Python 2 relies on Environment variable in PATH - attempt to use as follows
    except AttributeError:
        phantomjs_path = "phantomjs"

    try:
        proc = Popen([phantomjs_path, "--version"], stdout=PIPE, stderr=PIPE)
        proc.wait()

    except OSError:
        raise RuntimeError('PhantomJS is not present in PATH. Try "conda install phantomjs" or \
            "npm install -g phantomjs-prebuilt"')
    return phantomjs_path
       

### from bokeh/io, slightly modified to avoid their import_required util
### - https://github.com/bokeh/bokeh/blob/master/bokeh/io/export.py
def create_default_webdriver():
    '''Return phantomjs enabled webdriver'''
    phantomjs_path = detect_phantomjs()
    return webdriver.PhantomJS(executable_path=phantomjs_path, service_log_path=devnull)

create a plot¶

Here we create some data and save the plot as an html file for reading in

In [ ]:

df = pd.DataFrame(
    {'fruits': ['Apples', 'Pears', 'Nectarines', 'Plums', 'Grapes', 'Strawberries'],
     'counts': [5, 3, 4, 2, 4, 6] })

data = [go.Bar(x=df['fruits'],
               y=df['counts'])]

html_file = 'plotly-fruit-plot.html'

plotly.offline.plot(data, filename=html_file)

the magic¶

This, again, is stolen entirely from how bokeh is doing this. I really don't entirely understand it, for example, how it knows the perimeter of the plot as it's cropped so nicely with no modification from their call. I find that calling driver.maximize_window() give a bigger output file, but I needed to wait just a hair or it would grab the screenshot before this too effect. I didn't see any difference between using or omitting the driver.execute_script() call, so I left it commented.

In [46]:

### create webdrive, open file, maximize, and sleep
driver = create_default_webdriver()
driver.get(html_file)
driver.maximize_window()
#driver.execute_script("document.body.style.width = '100%';")
time.sleep(1)

fname = 'plotly-fruit-plot.png'

### original bokeh call gets byte data
### - http://selenium-python.readthedocs.io/api.html
# png = driver.get_screenshot_as_png()
# image = Image.open(io.BytesIO(png))

### I found this and seems simpler (could eliminate io import)
png = driver.get_screenshot_as_file(fname)
image = Image.open(fname)
driver.quit()

image

Out[46]: