Advanced web scrapper for machine learning and data science buit around BeautifulSoup and Pandas
Reason this release was yanked:
Code/Import changes
Project description
Introduction
Zineb is a lightweight tool solution for simple and efficient web scrapping and crawling built around BeautifulSoup and Pandas. It's main purpose is to help quickly structure your data in order to be used as fast as possible in data science or machine learning projects.
Understanding how Zineb works
Zineb gets your custom spider, creates a set of HTTPRequest
objects for each url, sends the requests and caches a BeautifulSoup object of the page within an HTMLResponse
class of that request.
Most of your interactions with the HTML page will be done through the HTMLResponse
class.
When the spider starts crawling the page, each response and request in past through the start function:
def start(self, response, **kwargs):
request = kwargs.get('request')
images = response.images
Getting started
Creating a project
To create a project do python -m zineb startproject <project name>
which will create a directory which will have the following structure.
.myproject | |--media | |-- models |-- base.py | |-- init.py | |-- manage.py | |-- settings.py | |-- spiders.py
Once the project folder is created, all your interractions with Zineb will be made trough the management commands that are executed through python manage.py
from your project's directory.
The models directory allows you to place the elements that will help structure the data that you have scrapped from from the internet.
The manage.py
file will allow you to run all the required commands from your project.
Finally, the spiders module will contain all the spiders for your project.
Configuring your project
On startup, Zineb implements a set of basic settings (zineb.settings.base
) that will get overrided by the values that you would have defined in your settings.py
located in your project.
You can read more about this in the settings section of this file.
Creating a spider
Creating a spider is extremely easy and requires a set of starting urls that can be used to scrap one or many HTML pages.
class Celebrities(Zineb):
start_urls = ['http://example.com']
def start(self, response, request=None, soup=None, **kwargs):
# Do something here
Once the Celibrities class is called, each request is passed through the start
method. In other words the zineb.http.responses.HTMLResponse
, zineb.http.request.HTTPRequest
and the BeautifulSoup
HTML page object are sent through the function.
You are not required to use all these parameters at once. They're just for convinience.
In which case, you can also write the start method as so if you only need one of these.
def start(self, response, **kwargs):
# Do something here
Other objects can be passes through the function such as the models that you have created but also the settings of the application etc.
Adding meta options
Meta options allows you to customize certain very specific behaviours [not found in the settings.py
file] related to the spider.
class Celerities(Zineb):
start_urls = ['http://example.com']
class Meta:
domains = []
Domains
This option limits a spider to a very specific set of domains.
Verbose name
This option writter as verbose_name
will specific a different name to your spider.
Running commands
Start
Triggers the execution of all the spiders present in the given the project. This command will be the main one that you will be using to execute your project.
Shell
Start a iPython shell on which you can test various elements on the HTML page.
When the shell is started, the zineb.http.HTTPRequest
, the zineb.response.HTMLResponse
, and the BeautifulSoup instance of the page are all injected in the shell.
Extractors are passed using aliases:
links
: LinkExtractorimages
: ImageExtractormultilinks
: MultiLinkExtractortables
: TableExtractor
The extractors are also all passed within the shell in addition to the project settings.
In that regards, the shell becomes a interesting place where you can test various querying on an HTML page before using it in your project. For example, using the shell with http://example.com.
We can get a simple url :
IPython 7.19.0
In [1]: response.find("a")
Out[1]: <a href="https://www.iana.org/domains/example">More information...</a>
We can find all urls on the page:
IPython 7.19.0
In [2]: extractor = links()
In [3]: extractor.resolve(response)
In [4]: str(extrator)
Out [4]: [Link(url=https://www.iana.org/domains/example, valid=True)]
In [5]: response.links
Out [5]: [Link(url=https://www.iana.org/domains/example, valid=True)]
Or simply get the page title:
IPython 7.19.0
In [6]: response.page_title
Out [6]: 'Example Domain'
Remember that in addition to the custom functions created for the class, all the rest called on zineb.response.HTMLResponse
are BeautifulSoup functions (find, find_all, find_next...)
Queries on the page
Like said previously, the majority of your interactions with the HTML page will be done through the HTMLResponse
object or zineb.http.responses.HTMLResponse
class.
This class will implement some very basic general functionnalities that you can use through the course of your project. To illustrate this, let's create a basic Zineb HTTP response from a request:
from zineb.http.requests import HTTPRequest
request = HTTPRequest("http://example.com")
Requests, when created a not sent [or resolved] automatically if the _send
function is not called. In that case, they are marked as being unresolved ex. HTTPRequest("http://example.co", resolved=False)
.
Once the _send
method is called, by using the html_page
attribute or calling any BeautifulSoup function on the class, you can do all the classic querying on the page e.g. find, find_all...
request._send()
request.html_response
-> Zineb HTMLResponse object
request.html_response.html_page
-> BeautifulSoup object
request.find("a")
-> BeautifulSoup Tag
If you do not know about BeautifulSoup please read the documentation here.
For instance, suppose you have a spider and want to get the first link present on the http://example.com page. That's what you would so:
from zineb.app import Zineb
class MySpider(Zineb):
start_urls = ["http://example.com"]
def start(self, response=None, request=None, soup=None, **kwargs):
link = response.find("a")
# Or, you can also use this tehnic through
# the request object
link = request.html_response.find("a")
# Or you can directly use the soup
# object as so
link = soup.find("a")
In order to understand what the Link
, Image
and Table
objects represents, please read the following section of this page.
Zineb HTTPRequest objects are better explained in the following section.
Getting all the links
request.html_response.links
-> [Link(url=http://example.com valid=True)]
Getting all the images
request.html_response.images
-> [Image(url=https://example.com/1.jpg")]
Getting all the tables
request.html_response.tables
-> [Table(url=https://example.com/1")]
Getting all the text
Finally you can retrieve all the text of the web page at once.
request.html_response.text
-> '\n\n\nExample Domain\n\n\n\n\n\n\n\nExample Domain\nThis domain is for use in illustrative examples in documents. You may use this\n domain in literature without prior coordination or asking for permission.\nMore information...\n\n\n\n'
Models
Models are a simple way to structure your scrapped data before saving them to a file. The Model class is built around Panda's excellent DataFrame class in order to simplify as a much as possible the fact of dealing with your data.
Creating a custom Model
In order to create a model, subclass the Model object from zineb.models.Model
and then add fields to it. For example:
from zineb.models.datastructure import Model
from zineb.models.fields import CharField
class Player(Model):
name = CharField()
Using the custom model
On its own, a model does nothing. In order to make it work, you have to add values to it and then resolve the fields.
You can add values to your model in two main ways.
Adding a free custom value
The first method consists of adding values through the add_value
method. This method does not rely on the BeautifulSoup HTML page object which means that values can be added freely.
player.add_value('name', 'Kendall Jenner')
Adding an expression based value
Addind expression based values requires a BeautifulSoup HTML page object. You can add one value at a time or multiple values.
player.add_expression("name", "a#kendall__text", many=True)
By using the many
parameter, you can add the all the tags with a specific name and/or attributes to your model at once.
Here is a list of expressions that you can use for this field:
expression | interpretation | tag |
---|---|---|
a.kendall | Link with class kendall | |
a#kendall | Lind with ID Kendall |
By default, if a pseudo is not provided, __text
pseudo is appended in order to always retrieve the inner text element of the tag.
Meta options
By adding a Meta to your model, you can pass custom behaviours.
- Ordering
- Indexing
Indexes
Ordering
Fields
Fields are a very simple way to passing HTML data to your model in a very structured way. Zineb comes with number of preset fields that you can use out of the box:
- CharField
- TextField
- NameField
- EmailField
- UrlField
- ImageField
- IntegerField
- DecimalField
- DateField
- AgeField
- FunctionField
- ArrayField
- CommaSeparatedField
How fields work
Once the field is called via the resolve
function on each field which in turns calls the super().resolve
function of the Field
super class, the value is stored.
By default, the resolve function will do the following things.
First, it will run all cleaning functions on the value for example by stripping tags like "<" or ">" by using the w3lib.html.remove_tags
library.
Second, a deep_clean
method will be called on the value which takes out any spaces using w3lib.html.strip_html5_whitespace
, remove escape characters with the w3lib.html.replace_escape_chars
function and finally reconstruct the value to ensure that any none-detected white space be eliminated.
Finally, all validators (default and custom created) are called on the value. The final value is then returned within the model class.
CharField
The CharField represents the normal character element on an HTML page. You constrain the length.
TextField
The text field is longer which allows you then to add paragraphs of text.
NameField
The name field allows to implement names in your model. The title
method is called on the string in order to represent the value correctly e.g. Kendall Jenner.
EmailField
The email field represents emails. The default validator, validators.validate_email
, is automatically called on the resolve function fo the class in order to ensure that that the value is indeed an email.
limit_to_domains
: Check if email corresponds to the list of specified domains
UrlField
The url field is specific for urls. Just like the email field, the default validator, validators.validate_url
is called in order to validate the url.
ImageField
The image field holds the url of an image exactly like the UrlField with the sole difference that you can download the image directly when the field is evaluated.
download
: Download the image to your media folder while the scrapping is performedas_thumnail
: Download image as a thumbnaildownload_to
: Download image to a specific path
class MyModel(Model):
avatar = ImageField(download=True, download_to="/this/path")
IntegerField
This field allows you to pass an integer into your model.
default
: Default value if Nonemax_value
: Implements a maximum value constraintmin_value
: Implements a minimum value constraint
DecimalField
This field allows you to pass a float value into your model.
default
: Default value if Nonemax_value
: Implements a maximum value constraintmin_value
: Implements a minimum value constraint
DateField
The date field allows you to pass dates to your model. In order to use this field, you have to pass a date format so that the field can know how to resolve the value.
date_format
: Indicates how to parse the incoming data valuedefault
: Default value if Nonetz_info
: Timezone information
class MyModel(Model):
date = DateField("%d-%m-%Y")
AgeField
The age field works likes the DateField but instead of returning the date, it will return the difference between the date and the current date which corresponds to the age.
date_format
: Indicates how to parse the incoming data valuedefault
: Default value if Nonetz_info
: Timezone information
FunctionField
The function field is a special field that you can use when you have a set of functions to run on the value before returning the final result. For example, let's say you have this value Kendall J. Jenner
and you want to run a specific function that takes out the middle letter on every incoming values:
def strip_middle_letter(value):
# Do something here
return value
class MyModel(Model):
name = FunctionField(strip_middle_letter, output_field=CharField())
Every time the resolve function will be called on this field, the methods provided will be passed on the value sequentially. Each method should return the new value.
An output field is not compulsory but if not provided, each value will be returned as a character.
ArrayField
An array field will store an array of values that are all evalutated to an output field that you would have specified.
N.B. Note that the value of an ArrayField is implemented as is in the final DataFrame. Make sure you are using this field correctly in order to avoid unwanted results.
CommaSeperatedField
Create a comma separated field in your model.
N.B. Note that the value of a CommaSeperatedField is implemented as is in the final DataFrame. Make sure you are using this field correctly in order to avoid unwanted results.
Creating your own field
You an also create a custom field by suclassing zineb.models.fields.Field
. When doing so, your custom field has to provide a resolve
function in order to determine how the value should be treated. For example:
class MyCustomField(Field):
def resolve(self, value):
initial_result = super().resolve(value)
# Rest of your code here
If you want to use the cleaning functionalities from the super class in your own resolve function, make sure to call super beforehand as indicated above.
Validators [initial validators]
Validators make sure that the value that was passed respects the constraints that were implemented as a keyword arguments on the field class. There are five basic validations that could possibly run if you specify a constraint for them:
- Maximum length (
max_length
) - Nullity (
null
) - Defaultness (
default
) - Validity (
validators
)
Maximum or Minimum length
The maximum or minimum length check ensures that the value does not exceed a certain length using zineb.models.validators.max_length_validator
or zineb.models.validators.min_length_validator
which are encapsulated and used within the zineb.models.validators.MinLengthValidator
or zineb.models.validators.MaxLengthValidator
class.
Nullity
The nullity validation ensures that the value is not null and that if a default is provided, that null value be replaced by the latter. It uses zineb.models.validators.validate_is_not_null
.
The defaultness provides a default value for null or none existing ones.
Practical examples
For instance, suppose you want only values that do not exceed a certain length:
name = CharField(max_length=50)
Or suppose you want a default value for fields that are empty or blank:
name = CharField(default='Kylie Jenner')
Remember that validators will validate the value itself for example by making sure that an URL is indeed an url or that an email follows the expected pattern that you would expect from an email.
Suppose you want only values that would be Kendall Jenner
. Then you could create a custom validator that would do the following:
def check_name(value):
if value == "Kylie Jenner":
return None
return value
name = CharField(validators=[check_name])
You can also create validators that match a specific regex pattern using the zineb.models.validators.regex_compiler
decorator:
from zineb.models.datastructure import Model
from zineb.models.fields import CharField
from zineb.models.validators import regex_compiler
@regex_compiler(r'\d+')
def custom_validator(value):
if value > 10:
return value
return 0
class Player(Model):
age = IntegerField(validators=[custom_validator])
IN this specific situation, it is important to understand that the result of the regex compiler is reinjected into your custom validator on which you can then do various other checks.
Field resolution
In order to get the complete structured data, you need to call resolve_values
which will return a pandas.DataFrame
object:
player.add_value("name", "Kendall Jenner")
player.resolve_values()
-> pandas.DataFrame
Practically though, you'll be using the save
method which also calls the resolve_values
under the hood:
player.save(commit=True, filename=None, **kwargs)
-> pandas.DataFrame or new file
By calling the save method, you'll be able to store the data directly to a JSON or CSV file.
Extractors
Extractors are utilities that facilitates extracting certain specific pieces of data from a web page such as links, images [...] quickly. They can be found in zineb.extactors
.
Some extractors can be used in various manners. First, with a context processor:
extractor = LinkExtractor()
with extractor:
# Do something here
Second, in an interation process:
for link in extractor:
# Do something here
Finally, with next
:
next(extractor)
You can also check if an extractor has a specific value and even concatenate some of them together:
# Contains
if x in extractor:
# Do something here
# Addition
concatenated_extractors = extractor1 + extractor2
LinkExtractor
url_must_contain
- only keep urls that contain a specific stringunique
- return a unique set of urls (no duplicates)base_url
- reconcile a domain to a pathonly_valid_links
- only keep links (Link) that are marked as valid
extractor = LinkExtractor()
extractor.finalize(response.html_response)
-> [Link(url=http://example.com, valid=True)]
There might be times where the extracted links are relative paths. This can cause an issue for running additional requests. In which case, use the base_url
parameter:
extractor = LinkExtractor(base_url=http://example.com)
extractor.finalize(response.html_response)
# Instead of getting this result which would also
# be marked as a none valid link
-> [Link(url=/relative/path, valid=False)]
# You will get the following with the full url link
-> [Link(url=http://example.com/relative/path, valid=True)]
NOTE: By definition, a relative path is not a valid link hence the valid set to False.
MultiLinkExtractor
A MultiLinkExtractor
works exactly like the LinkExtractor
with the only difference being that it also identifies and collects emails that are contained within the HTML page.
TableExtractor
Extract all the rows from the first table that is matched on the HTML page.
class_name
- intercept a table with a specific class namehas_headers
- specify if the table has headers in order to ignore it in the final datafilter_empty_rows
- ignore any rows that do not have a valuesprocessors
- a set of functions to run on the data once it is all extracted
ImageExtractor
Extract all the images on the HTML page.
You can filter down the images that you get by using a specific set of parameters:
unique
- return only a unique et set of urlsas_type
- only return images having a specific extensionurl_must_contain
- only return images which contains a specific stringmatch_height
- only return images that match as specific heightmatch_width
- only return images that match a specific width
TextExtractor
Extract all the text on an HTML page.
First, the text is retrieved as a raw value then tokenized and vectorized using nltk.tokenize.PunktSentenceTokenizer
and nltk.tokenize.WordPunctTokenizer
.
To know more about NLKT, please read the following documentation.
Zineb special wrappers
HTTPRequest
Zineb uses a special built-in HTTPRequest class which wraps the following for better cohesion:
- The
requests.Request
response class - The
bs4.BeautifulSoup
object
In general, you will not need to interact with this class that much because it's just an interface for implement additional functionnalities especially to the Request class.
follow
: create a new instance of the class whose resposne will be the one of a new urlfollow_all
: create new instances of the class who responses will tbe the ones of the new urlsurljoin
: join a path the domain
HTMLResponse
It wraps the BeautifulSoup object in order to implement some small additional functionalities:
page_title
: return the page's titlelinks
: return all the links of the pageimages
: return all the images of the pagetables
: return all the tables of the page
Signals
Signals are a very simple yet efficient way for you to run functions during the lifecycle of your project when certain events occur at very specific moments.
Internally signals are sent on the following events:
- When the registry is populated
- Before the spider starts
- After the spider has started
- Before an HTTP request is sent
- Before and HTTP request is sent
- Before the model downloads anything
- After the model has downloaded something
Creating a custom signal
To create custom signal, you need to mark a method as being a receiver for any incoming signals. For example, if you want to create a signal to intercept one of the events above, you should do:
from zineb.signals import receiver
@receiver(tag="Signal Name")
def my_custom_signal(sender, **kwargs):
pass
The signals function has to be able to accept a sender
object and additional parameters such as the current url or the current HTML page.
You custom signals do not have to return anything.
Pipelines
Pipelines are a great way to send chained requests to the internet or treat a set of responses by processing them afterwards through a set of functions of your choice.
Some Pipeplines are also perfect for donwloading images.
ResponsesPipeline
The response pipepline allows you to chain a group of responses and treat all of them at once through a function:
from zineb.http.pipelines import ResponsesPipeline
pipeline = ResponsesPipeline([response1, response2], [function1, function2])
pipeline.results
-> list
It comes with three main parameters:
responses
- which corresponds to a list of HTMLResponsesfunctions
- a list of functions to pass each individual response and additional parametersparamaters
- a set of additional parameters to pass to the functions
The best way to use the ResponsesPipeline is within the functions of your custom spider:
class MySpider(Zineb):
start_urls = ["https://example.com"]
def start(self, response, soup=None, **kwargs):
extractor = LinksExtractor()
extractor.resolve(soup)
responses = request.follow_all(*list(extractor))
ResponsesPipeline(responses, [self.do_something_here])
def do_something_here(self, response, soup=None, **kwargs):
# Continue parsing data here
N.B. Each function is executed sequentially. So, the final result will come from the final function of the list
HTTPPipeline
This pipeline takes a set of urls, creates HTTPResquests for each of them and then sends them to the internet.
If you provided a set of functions, it will pass each request through them.
from zineb.http.pipelines import HTTPPipeline
from zineb.utils.general import download_image
HTTPPipeline([https://example.com], [download_image])
Each function should be able to accept an HTTP Response object.
You can also pass additional parameters to your functions by doing the following:
HTTPPipeline([https://example.com], [download_image], parameters={'extra': False})
In this specific case, your function should accept an extra
parameter which result would be False.
Callback
The Callback class allows you to run a callback function once each url is processed and passed through the main start function of your spider.
The __call__
method is triggerd on the instance in order to resolve the function to use.
class Spider(Zineb):
start_urls = ["https://example.com"]
def start(self, response, **kwargs):
request = kwargs.get("request")
model = MyModel()
return Callback(request.follow, self.another_function, model=model)
def another_function(self, response, **kwargs):
model = kwargs.get("model")
model.add_value("name", "Kendall Jenner")
model.save()
Utilities
Link reconciliation
Most of times, when you retrieve links from a page, they are returned as relative paths. The urljoin
method reconciles the url of the visited page with that path.
<a href="/kendall-jenner">Kendall Jenner</a>
# Now we want to reconcile the relative path from this link to
# the main url that we are visiting e.g. https://example.com
request.urljoin("/kendall-jenner")
-> https://example.com/kendall-jenner
Settings
This section will talk about all the available settings that are available for your project and how to use them for web scrapping.
PROJECT_PATH
Represents the current path for your project. This setting is not be changed.
SPIDERS
In order for your spider to be executed, every created spider should be registered here. The name of the class should serve as the name of the spider to be used.
SPIDERS = [
"MySpider"
]
DOMAINS
You can restrict your project to use only to a specific set of domains by ensuring that no request is sent if it matches one of the domains within this list.
DOMAINS = [
"example.com"
]
ENSURE_HTTPS
Enforce that every link in your project is a secured HTTPS link. This setting is set to False by default.
MIDDLEWARES
Middlewares are functions/classes that are executed when a signal is sent from any part of the project. Middlewares implement extra functionnalities without affecting the core parts of the project. They can then be disabled safely if you do not need them.
MIDDLEWARES = [
"zineb.middlewares.handlers.Handler",
"myproject.middlewares.MyMiddleware"
]
The main Zineb middlewares are the following:
- zineb.middlewares.referer.Referer
- zineb.middlewares.handlers.Handler
- zineb.middlewares.automation.Automation
- zineb.middlewares.history.History
- zineb.middlewares.statistics.GeneralStatistics
- zineb.middlewares.wireframe.WireFrame
USER_AGENTS
A user agent is a characteristic string that lets servers and network peers identify the application, operating system, vendor, and/or version of the requesting MDN Web Docs.
Implement additional sets of user agents to your projects in addition to those that were already created.
RANDOMIZE_USER_AGENTS
Specifies whether to use one user agent for every request or to randomize user agents on every request. This setting is set to to False by default.
DEFAULT_REQUEST_HEADERS
Specify additional default headers to use for each requests.
The default initial headers are:
Accept-Language
- enAccept
- text/html,application/json,application/xhtml+xml,application/xml;q=0.9,/;q=0.8- Referrer - None
PROXIES
Use a set of proxies for each request. When a request in sent, a random proxy is selected and implemented with the request.
PROXIES = [
("http", "127.0.0.1"),
("https", "127.0.0.1")
]
RETRY
Specifies the retry policy. This is set to False by default. In other words, the request silently fails and never retries.
RETRY_TIMES
Specificies the amount of times the the request is sent before eventually failing.
RETRY_HTTP_CODES
Indicates which status codes should trigger a retry. By default, the following codes: 500, 502, 503, 504, 522, 524, 408 and 429 will trigger it.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.