pip install bluemossfrom bluemoss import Node, scrape
node = Node('a') # extract the text contained in the first a-tag
scrape(node, YOUR_HTML)With bluemoss you create a Node object to scrape data from any website.
That Node object is like a recipe 👩‍🍳 which describes what to scrape from which tags,
and how to transform and structure the scraped data into the format you need.
bluemoss uses XPath 1.0 to locate html tags. If you are new to XPath, no problem — ChatGPT
has got your back.
This section will show you how can to scrape the web with bluemoss.
For all examples that follow, we are going to scrape the html document below.
<html>
<body>
<li>
<a href="/portfolio?company=apple">
Apple
</a>
<div class="location_us">
<p>Cupertino</p>
</div>
</li>
<li>
<a href="/portfolio?company=google">
Google
</a>
<div class="location_us">
<p>Mountain View</p>
</div>
</li>
<li>
<a href="/portfolio?company=deepmind">
DeepMind
</a>
<div class="location_uk">
<p>London</p>
</div>
</li>
</body>
</html>Example 1 - Scrape the text in the first a tag.
The Node object below instructs the scrape function to find the first a-tag and extract the text it contains.
from bluemoss import Node, scrape
node = Node('a')
scrape(node, HTML) == 'Apple'Example 2 - Scrape the second company headquarters which are located in the US
node = Node('div[contains(@class, "location_us")]', filter=1)
scrape(node, HTML) == 'Mountain View'Pro Tip: Learning XPath is easy and fast with ChatGPT.
This example introduces the filter argument which determines which of those tag(s) that match the xpath locator will be scraped. The default value for the filter arg is 0 (index 0), which will scrape the very first tag that matches the given xpath. Our goal for this example was to extract the text of the second tag (index 1) that matches our xpath, therefor we set filter = 1.
Example 3 - Scrape ALL company headquarters which are located in the US
node = Node('div[contains(@class, "location_us")]', filter=None)
scrape(node, HTML) == ['Cupertino', 'Mountain View']Setting filter=None will filter for all tags matched against the given xpath.
Example 4 - Scrape the first and third company names.
Node('a', filter=[0, 2])
scrape(node, HTML) == ['Apple', 'DeepMind']In this example we set the filter arg to a list of ints. Those int values refer to the first and third index (0 and 2).
Example 5 - Scrape all company names, but exclude the first one.
The expected scrape result is
['Google', 'DeepMind']
# 'Apple' not included, since it is located in the first a-tag (index 0)Below are a couple of valid solutions, and the first one is the most recommended one, since it keeps the arguments provided to the Node object as simple as possible:
# solution 1, the recommended solution
from bluemoss import Node, Range
Node('a', filter=Range(1))
# Range(1) filters the matched tags from index 1 onwards# solution 2
Node('li//a', filter=Range(1))
# the xpaths 'a' and 'li//a' match the same tags in our html doc# solution 3
Node('a', filter=[1, 2])# solution 4
Node('a', filter=Range(1, 3))
# The Range class accepts a second int argument (the stop index).
# Range(x, y) filters for the matched tags at indices x, ..., y-1
# e.g. Range(2, 6) filters for indices 2, 3, 4, 5.Example 6 - Scrape all company names from index 1 onwards in reverse order.
The expected scrape result is
['DeepMind', 'Google']# solution 1
Node('a', filter=Range(1, reverse=True)) # set reverse to True# solution 2
Node('a', filter=Range(1), transform=lambda res: res[::-1]) # use transform argThe first two examples simply set the reverse arg of the Range object to True. The last example uses a different approach. It introduces the transform arg of the Node class.
The transform function is the function being executed after
- the tags were matched with the given xpath
- the matched tags were further filtered using the Node.filter arg
Pro Tip: The transform-function defines the last step in computing the scrape-result of a Node object.
Example 7 - Scrape the last two company names and store the result in a dict under the key 'companies'
Node('a', filter=[-2, -1], key='companies')
scrape(node, HTML) == {'companies': ['Google', 'DeepMind']}In the example above we provide the indexes -2 and -1 to the filter-list as those indices represent the last two elements in a Python list.
Example 8 - Scrape the first company name and store the result in a dict under the key 'companies'
Node('a', key='companies')
scrape(node, HTML) == {'companies': 'Apple'}Example 9 - Scrape the first company-id.
Let us first define a helper function that will receive the href-value as an argument and return the company id:
# helper function to the solutions
def get_company_id(href: str) -> str:
return href.split('=')[-1]# solution 1
from src.bluemoss import Node, Ex
Node('a', extract=Ex.HREF, transform=get_company_id)
# Declare the tag-property to be extracted by using the 'extract' arg.
# Introducing the 'Ex' Enum which provides handy types of extraction.# solution 2
Node('a', extract='href', transform=get_company_id)
# The 'extract' arg also accept string values.# solution 3
Node('a/@href', transform=get_company_id)
# We can also just use xpath to extract the href property.Example 10 - Advanced Scraping - Scrape the name and headquarters of every company.
The expected result:
[
['Apple', 'Cupertino'],
['Google', 'Mountain View'],
['DeepMind', 'London']
]# solution
Node(
'li', # match 'li' tags
filter=None, # scrape all the matched 'li' tags
nodes=[
Node('a'), # within the 'li' tag, match the first 'a' tag and extract the text
Node('p') # within the 'li' tag, match the first 'p' tag and extract the text
]
)Pro Tip: The **nodes** arg let's you scrape multiple different tags within the same parent tag.
Example 11 - Advanced Scraping - Scrape the name and headquarters of every company. The result shall be a list of dicts.
The expected result:
[
{'name': 'Apple', 'location': 'Cupertino'},
{'name': 'Google', 'location': 'Mountain View'},
{'name': 'DeepMind', 'location': 'London'}
]# solution
Node(
'li',
filter=None,
nodes=[
Node('a', key='name'),
Node('p', key='location')
]
)Pro Tip: All Nodes in a nodes list either define the 'key' arg or none of them do.
Example 12 - Advanced Scraping - In this last example, we want to scrape the name and location of every company, as well as the total amount of companies located in the US and UK. We also want to store the scraped data not in a dict or list as we did in the previous examples, but instead want to store the data in dataclass instances.
Info - The code snippet below shows that we assume a dataclass called Companies in which we will store the entire scrape-result. The expected result also assumes, that the Companies instance exposes two properties dict and json
# expected result
companies: Companies = scrape(node, HTML)
companies.dict == {
'companies': [
{'name': 'Apple', 'location': 'Cupertino'},
{'name': 'Google', 'location': 'Mountain View'},
{'name': 'DeepMind', 'location': 'London'},
],
'amount_uk_companies': 1,
'amount_us_companies': 2,
'amount_companies': 3
}
companies.json == """{
"companies": [
{
"name": "Apple",
"location": "Cupertino"
},
{
"name": "Google",
"location": "Mountain View"
},
{
"name": "DeepMind",
"location": "London"
}
],
"amount_uk_companies": 1,
"amount_us_companies": 2,
"amount_companies": 3
}"""# solution
from dataclasses import dataclass
from bluemoss import Node, Jsonify
@dataclass
class Company(Jsonify):
name: str
location: str
@dataclass
class Companies(Jsonify):
companies: list[Company]
amount_uk_companies: int
amount_us_companies: int
@property
def amount_companies(self) -> int:
return self.amount_us_companies + self.amount_uk_companies
@property
def dict(self):
return super().dict | {'amount_companies': self.amount_companies}
Node(
target=Companies,
nodes=[
Node(
'li',
filter=None, # makes sure to filter all matched 'li' tags
key='companies', # will store the scrape-result as a dict under the key 'companies'
target=Company, # the data scraped from every 'li' tag will be transformed into a 'Company' instance
nodes=[ # every 'Company' instance will be initialized with the parameters 'name' and 'location'
Node('a', key='name'), # extracts the text from the 'li/a' tag and stores it in the Company.name parameter
Node('p', key='location') # extracts the text from the 'li/p' tag and stores it in the Company.location parameter
]
),
Node(
"div[@class='location_uk']", # finds all 'div' tags whose class property contains the text 'location_uk'
filter=None, # filter all matched tags
key='amount_uk_companies', # stores the result in the Companies.amount_uk_companies parameter
transform=lambda companies: len(companies) # The initial result is a list (due to filter=None) and it
# will then be mapped to its length.
# Therefor the final scrape-result is an int.
),
Node(
"div[@class='location_us']",
filter=None,
key='amount_us_companies',
transform=lambda companies: len(companies)
)
]
)The solution in the code snippet introduces two new things:
- The Node.target parameter allows us to specify what class or dataclass to use in order to store the data scraped in the Node.nodes list. If a Node instance sets a target, then all instances in Node.nodes must set their key parameter and every key must map to one of the init parameters of Node.target.
- The Jsonify class exposes the two properties dict and json. It therefor makes it easy transform the scraped data stored in dataclass instances into a Python dict or json string.
Pro Tip: The Jsonify class will exclude any parameters starting with an underscore "_" from appearing in the return of the .dict and .json properties. This enables us to hide certain parameters from appearing in those returns.
- type safety - Dataclass instances as used in this example enforce typed parameters.
- properties - Sometimes we want our data transformations to take place inside the dataclass, e.g. through properties. Properties provide a simple way to derive data from the instance parameters of a class instance. By moving the data transformation step from the Node.transform parameter to a dataclass property, we make the transformation explicitly available to the dataclass.
- post_init - The post_init method that is available in Python dataclasses is yet another nice step to manipulate the instance parameters and therefore move the data transformation step partially or as a whole from the Node.transform parameter to the post_init method of the dataclass.
- MacOS
- Linux
- Windows
- 3.9
- 3.10
- 3.11
- 3.12
Apache 2.0