Of course, we won’t be able to cover all aspect of every tool we discuss. But this post should be enough to get you a good idea of which tools do what, and when to use each.

Note: when I talk about Python in this blog post you should assume that I’m talking about Python3.

Summary:

• Web Fundamentals
• Manually opening a socket and sending the HTTP request
• urllib3 & LXML
• requests & BeautifulSoup
• Scrapy
• Selenium & Chrome —headless
• Conclusion

Web Fundamentals

The internet is really complex: there are many underlying technologies and concepts involved to view a simple web page in your browser. I’m not going to explain everything, but I will show you the most important things you have to understand in order to extract data from the web.

HyperText Transfer Protocol

HTTP uses a client/server model, where an HTTP client (A browser, your Python program, curl, Requests, and so on) opens a connection and sends a message (“I want to see that page: /product”) to an HTTP server (like Nginx, Apache, and others).

Then the server answers with a response (the HTML code, for example) and closes the connection. HTTP is called a stateless protocol, because each transaction (request/response) is independent. FTP, for example, is stateful.

Basically, when you type a website address in your browser, the HTTP request looks like this:

import socket


HOST = 'www.google.com'  # Server hostname or IP address
PORT = 80        # Port


client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server_address = (HOST, PORT)
client_socket.connect(server_address)


request_header = b'GET / HTTP/1.0\r\nHost: www.google.com\r\n\r\n'
client_socket.sendall(request_header)


response = ''
while True:
    recv = client_socket.recv(1024)
    if not recv:
        break
    response += str(recv)


print(response)
client_socket.close()  

In the first line of this request, you can see multiples things:

• Web Fundamentals
• Manually opening a socket and sending the HTTP request
• urllib3 & LXML
• requests & BeautifulSoup
• Scrapy
• Selenium & Chrome —headless
• Conclusion

Here are the most important header fields:

• Web Fundamentals
• Manually opening a socket and sending the HTTP request
• urllib3 & LXML
• requests & BeautifulSoup
• Scrapy
• Selenium & Chrome —headless
• Conclusion

And the list goes on…you can find the full header list here.

A server will respond with something like this:

GET /product/ HTTP/1.1
Host: example.com
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/web\
p,*/*;q=0.8
Accept-Encoding: gzip, deflate, sdch, br
Connection: keep-alive
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_11_6) AppleWebKit\
/537.36 (KHTML, like Gecko) Chrome/56.0.2924.87 Safari/537.36

On the first line, we have a new piece of information, the HTTP code 200 OK. It means the request has succeeded. As for the request headers, there are lots of HTTP codes, split into four common classes, 2XX for successful requests, 3XX for redirects, 4XX for bad requests (the most famous being 404 Not found), and 5XX for server errors.

Then, in case you are sending this HTTP request with your web browser, the browser will parse the HTML code, fetch all the eventual assets (Javascript files, CSS files, images…) and it will render the result into the main window.

In the next parts we will see the different ways to perform HTTP requests with Python and extract the data we want from the responses.

Manually opening a socket and sending the HTTP request

Socket

The most basic way to perform an HTTP request in Python is to open a socket and manually send the HTTP request.

import socket


HOST = 'www.google.com'  # Server hostname or IP address
PORT = 80        # Port


client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server_address = (HOST, PORT)
client_socket.connect(server_address)


request_header = b'GET / HTTP/1.0\r\nHost: www.google.com\r\n\r\n'
client_socket.sendall(request_header)


response = ''
while True:
    recv = client_socket.recv(1024)
    if not recv:
        break
    response += str(recv)


print(response)
client_socket.close()

Now that we have the HTTP response, the most basic way to extract data from it is to use regular expressions.

Regular Expressions

A regular expression (RE, or Regex) is a search pattern for strings. With regex, you can search for a particular character/word inside a bigger body of text.

For example, you could identify all phone numbers inside a web page. You can also replace items, for example, you could replace all uppercase tag in a poorly formatted HTML by lowercase ones. You can also validate some inputs …

The pattern used by the regex is applied from left to right. Each source character is only used once. You may be wondering why it is important to know about regular expressions when doing web scraping?

After all, there is all kind of different Python module to parse HTML, with XPath, CSS selectors.

In an ideal semantic world, data is easily machine-readable, the information is embedded inside relevant HTML element, with meaningful attributes.

But the real world is messy, you will often find huge amounts of text inside a p element. When you want to extract a specific data inside this huge text, for example, a price, a date, a name… you will have to use regular expressions.

Note: Here is a great website to test your regex: https://regex101.com/ and one awesome blog to learn more about them, this post will only cover a small fraction of what you can do with regexp.

Regular expressions can be useful when you have this kind of data:

<p>Price : 19.99$</p>

We could select this text node with an Xpath expression, and then use this kind of regex to extract the price :

^Price\s:\s(\d+\.\d{2})\$ 

To extract the text inside an HTML tag, it is annoying to use a regex, but doable:

import re


html_content = '<p>Price : 19.99$</p>'


m = re.match('<p>(.+)<\/p>', html_content)
if m:
	print(m.group(1))

As you can see, manually sending the HTTP request with a socket, and parsing the response with regular expression can be done, but it’s complicated and there are higher-level API that can make this task easier.

urllib3 & LXML

Disclaimer: It is easy to get lost in the urllib universe in Python. You have urllib and urllib2 that are parts of the standard lib. You can also find urllib3. urllib2 was split in multiple modules in Python 3, and urllib3 should not be a part of the standard lib anytime soon. This whole confusing thing will be the subject of a blog post by itself. In this part, I’ve made the choice to only talk about urllib3 as it is used widely in the Python world, by Pip and requests to name only them.

Urllib3 is a high-level package that allows you to do pretty much whatever you want with an HTTP request. It allows doing what we did above with socket with way fewer lines of code.

import urllib3
http = urllib3.PoolManager()
r = http.request('GET', 'http://www.google.com')
print(r.data)

Much more concise than the socket version. Not only that, but the API is straightforward and you can do many things easily, like adding HTTP headers, using a proxy, POSTing forms …

For example, had we decide to set some headers and to use a proxy, we would only have to do this.

import urllib3
user_agent_header = urllib3.make_headers(user_agent="<USER AGENT>")
pool = urllib3.ProxyManager(f'<PROXY IP>', headers=user_agent_header)
r = pool.request('GET', 'https://www.google.com/')

See? Exactly the same number of line, however, there are some things that urllib3 does not handle very easily, for example, if we want to add a cookie, we have to manually create the corresponding headers and add it to the request.

There are also things that urllib3 can do that requsts can’t, creation and management of pool and proxy pool, control of retry strategy for example.

To put in simply, urllib3 is between requests and socket in terms of abstraction, although way closer to requests than socket.

This time, to parse the response, we are going to use the lxml package and XPath expressions.

XPath

Xpath is a technology that uses path expressions to select nodes or node- sets in an XML document (or HTML document). As with the Document Object Model, Xpath is a W3C standard since 1999. Even if Xpath is not a programming language in itself, it allows you to write expression that can access directly to a specific node, or a specific node-set, without having to go through the entire HTML tree (or XML tree).

Think of XPath as regexp, but specifically for XML/HMTL.

To extract data from an HTML document with XPath we need 3 things:

• Web Fundamentals
• Manually opening a socket and sending the HTTP request
• urllib3 & LXML
• requests & BeautifulSoup
• Scrapy
• Selenium & Chrome —headless
• Conclusion

To begin we will use the HTML that we got thanks to urllib3, we just want to extract all the links from the Google homepage so we will use one simple XPath expression: //aand we will use LXML to run it. LXML is a fast and easy to use XML and HTML processing library that supports XPATH.

Installation:

pip install lxml 

Below is the code that comes just after the previous snippet:

from lxml import html


# We reuse the reponse from urllib3
data_string = r.data.decode('utf-8', errors='ignore')
# We instantiate a tree object from the HTML
tree = html.fromstring(data_string)
# We run the XPath against this HTML
# This returns an array of element
links = tree.xpath('//a')
for link in links:
    # For each element we can easily get back the URL
    print(link.get('href'))

And the output should look like this:

https://books.google.fr/bkshp?hl=fr&tab=wp
https://www.google.fr/shopping?hl=fr&source=og&tab=wf
https://www.blogger.com/?tab=wj
https://photos.google.com/?tab=wq&pageId=none
http://video.google.fr/?hl=fr&tab=wv
https://docs.google.com/document/?usp=docs_alc
...
https://www.google.fr/intl/fr/about/products?tab=wh

You have to keep in mind that this example is really really simple and doesn’t really show you how powerful XPath can be (note: this XPath expression should have been changed to //a/@href to avoid having to iterate on links to get their href ).

If you want to learn more about XPath you can read this good introduction. The LXML documentation is also well written and is a good starting point.

XPath expresions, like regexp, are really powerful and one of the fastest way to extract information from HTML, and like regexp, XPath can quickly become messy, hard to read and hard to maintain.

requests & BeautifulSoup

Requests is the king of python packages, with more than 11 000 000 downloads, it is the most widly used package for Python.

Installation:

pip install requests 

Making a request with Requests (no comment) is really easy:

import requests


r = requests.get('https://www.scrapingninja.co')
print(r.text)

With Requests it is easy to perform POST requests, handle cookies, query parameters…

Authentication to Hacker News

Let’s say we want to create a tool to automatically submit our blog post to Hacker news or any other forums, like Buffer. We would need to authenticate to those websites before posting our link. That’s what we are going to do with Requests and BeautifulSoup!

Here is the Hacker News login form and the associated DOM:

There are three <input> tags on this form, the first one has a type hidden with a name “goto” and the two others are the username and password.

If you submit the form inside your Chrome browser, you will see that there is a lot going on: a redirect and a cookie is being set. This cookie will be sent by Chrome on each subsequent request in order for the server to know that you are authenticated.

Doing this with Requests is easy, it will handle redirects automatically for us, and handling cookies can be done with the Session object.

The next thing we will need is BeautifulSoup, which is a Python library that will help us parse the HTML returned by the server, to find out if we are logged in or not.

Installation:

pip install beautifulsoup4 

So all we have to do is to POST these three inputs with our credentials to the /login endpoint and check for the presence of an element that is only displayed once logged in:

import requests
from bs4 import BeautifulSoup


BASE_URL = 'https://news.ycombinator.com'
USERNAME = ""
PASSWORD = ""


s = requests.Session()


data = {"gogo": "news", "acct": USERNAME, "pw": PASSWORD}
r = s.post(f'{BASE_URL}/login', data=data)


soup = BeautifulSoup(r.text, 'html.parser')
if soup.find(id='logout') is not None:
	print('Successfuly logged in')
else:
	print('Authentication Error')

In order to learn more about BeautifulSoup we could try to extract every links on the homepage.

By the way, Hacker News offers a powerful API, so we’re doing this as an example, but you should use the API instead of scraping it!

The first thing we need to do is to inspect the Hacker News’s home page to understand the structure and the different CSS classes that we will have to select:

We can see that all posts are inside a <tr class="athing"> **so the first thing we will need to do is to select all these tags. This can be easily done with:

links = soup.findAll('tr', class_='athing') 

Then for each link, we will extract its id, title, url and rank:

import requests
from bs4 import BeautifulSoup


r = requests.get('https://news.ycombinator.com')
soup = BeautifulSoup(r.text, 'html.parser')
links = soup.findAll('tr', class_='athing')


formatted_links = []


for link in links:
	data = {
		'id': link['id'],
		'title': link.find_all('td')[2].a.text,
		"url": link.find_all('td')[2].a['href'],
		"rank": int(links[0].td.span.text.replace('.', ''))
	}
	formatted_links.append(data)


print(formatted_links)

As you saw, Requests and BeautifulSoup are great libraries to extract data and automate different things by posting forms. If you want to do large-scale web scraping projects, you could still use Requests, but you would need to handle lots of things yourself.

When you need to scrape a lots of webpages, there are many things you have to take care of:

• Web Fundamentals
• Manually opening a socket and sending the HTTP request
• urllib3 & LXML
• requests & BeautifulSoup
• Scrapy
• Selenium & Chrome —headless
• Conclusion

Fortunately for us, tools exist that can handle those things for us.

Scrapy

Scrapy is a powerful Python web scraping framework. It provides many features to download web pages asynchronously, process and save it. It handles multithreading, crawling (the process of going from links to links to find every URLs in a website), sitemap crawling and many more.

Scrapy has also an interactive mode called the Scrapy Shell. With Scrapy Shell you can test your scraping code really quickly, like XPath expression or CSS selectors.

The downside of Scrapy is that the learning curve is steep, there is a lot to learn.

To follow up on our example about Hacker news, we are going to write a Scrapy Spider that scrapes the first 15 pages of results, and saves everything in a CSV file.

You can easily install Scrapy with pip:

pip install Scrapy 

Then you can use the scrapy cli to generate the boilerplate code for our project:

scrapy startproject hacker_news_scraper 

Inside hacker*news*scraper/spider we will create a new python file with our Spider’s code:

from bs4 import BeautifulSoup
import scrapy




class HnSpider(scrapy.Spider):
    name = "hacker-news"
    allowed_domains = ["news.ycombinator.com"]
    start_urls = [f'https://news.ycombinator.com/news?p={i}' for i in range(1,16)]


    def parse(self, response):
        soup = BeautifulSoup(response.text, 'html.parser')
        links = soup.findAll('tr', class_='athing')


        for link in links:
        	yield {
        		'id': link['id'],
        		'title': link.find_all('td')[2].a.text,
        		"url": link.find_all('td')[2].a['href'],
        		"rank": int(link.td.span.text.replace('.', ''))
        	}

There is a lot of convention in Scrapy, here we define an Array of starting urls. The attribute name will be used to call our Spider with the Scrapy command line.

The parse method will be called on each URL in the start_urls array

We then need to tune Scrapy a little bit in order for our Spider to behave nicely against the target website.

# Enable and configure the AutoThrottle extension (disabled by default)
# See https://doc.scrapy.org/en/latest/topics/autothrottle.html
AUTOTHROTTLE_ENABLED = True
# The initial download delay
AUTOTHROTTLE_START_DELAY = 5

You should always turn this on, it will make sure the target website is not slow down by your spiders by analyzing the response time and adapting the numbers of concurrent threads.

You can run this code with the Scrapy CLI and with different output format (CSV, JSON, XML…):

scrapy crawl hacker-news -o links.json 

And that’s it! You will now have all your links in a nicely formatted JSON file.

Selenium & Chrome —headless

Scrapy is really nice for large-scale web scraping tasks, but it is not enough if you need to scrape Single Page Application written with Javascript frameworks because It won’t be able to render the Javascript code.

It can be challenging to scrape these SPAs because there are often lots of AJAX calls and websockets connections involved. If performance is an issue, you should always try to reproduce the Javascript code, meaning manually inspecting all the network calls with your browser inspector, and replicating the AJAX calls containing the interesting data.

In some cases, there are just too many asynchronous HTTP calls involved to get the data you want and it can be easier to just render the page in a headless browser.

Another great use case would be to take a screenshot of a page, and this is what we are going to do with the Hacker News homepage (again !)

You can install the selenium package with pip:

pip install selenium 

You will also need Chromedriver:

brew install chromedriver 

Then we just have to import the Webdriver from selenium package, configure Chrome with headless=True and set a window size (otherwise it is really small):

from selenium import webdriver
from selenium.webdriver.chrome.options import Options


options = Options()
options.headless = True
options.add_argument("--window-size=1920,1200")
driver = webdriver.Chrome(options=options, executable_path=r'/usr/local/bin/chromedriver')
driver.get("https://news.ycombinator.com/")
driver.save_screenshot('hn_homepage.png')
driver.quit()

You should get a nice screenshot of the homepage:

You can do many more with the Selenium API and Chrome, like :

• Web Fundamentals
• Manually opening a socket and sending the HTTP request
• urllib3 & LXML
• requests & BeautifulSoup
• Scrapy
• Selenium & Chrome —headless
• Conclusion

Selenium and Chrome in headless mode is really the ultimate combination to scrape anything you want. You can automate anything that you could do with your regular Chrome browser.

The big drawback is that Chrome needs lots of memory / CPU power. With some fine-tuning you can reduce the memory footprint to 300-400mb per Chrome instance, but you still need 1 CPU core per instance.

If you want to run several Chrome instances concurrently, you will need powerful servers (the cost goes up quickly) and constant monitoring of resources.

Conclusion

Here is a quick recap table of every technology we discussed in this article. Do not hesitate to let me know in the comments if you know some resources that you feel have their place here.

I hope that this overview will help you choose your Python scraping tools and that you learned something reading this post.

Everything I talked about in this post are things I used to build my new indie hacker project: ScrapingNinja, the simplest web scraping API around there 😊.

Every tool I talked about in this post will be the subject of a specific blog post in the future where I’ll go deep into the details.

Do not hesitate to let me know in the comments what else you’d like to know about scraping. I’ll talk about it in my next post.

Recommended Reading:

☞ 5 Python Frameworks You Should Learn 2019

☞ Data Preprocessing in Python

☞ Deep Learning from Scratch and Using Tensorflow in Python

☞ Deploying a python-django application using docker

☞ Build your Python Extensions with Rust

☞ Rock, Paper, Scissors With Python

☞ Python Programming for Data Science and Machine Learning

☞ How to Build an E-commerce Website with Django and Python

#python