![[S]crape](_images/scrape_logo2.png)
- the Documentation¶
[S]crape is a tool developed to help researchers extract selective data from web publications. It is particularly useful for serial web publications which have similar structure over many issues. You interactively develop a selection and extraction set of commands, and run them across a series of issues, generating output (JSON or CSV).
Overview¶
Modes of Operation¶
You operate [S]crape in one of two ways:
- interactively;
- batch;
[S]crape starts and opens targets through a browser [1], and gets it’s data from that browser.
You can interactively highlight sections in the browser and scrape will give you unambiguous code to select that data. As with a Google search, skill will help make the search term (xpath or css selector) more general, yet specific enough to return your desired result.
Inspect results as you work interactively until you are satisfied and then save a range of commands from your history to develop your script.
At anytime you can load and run scripts against an opened target. Thus you build up a complete script incrementally. In batch mode, you can automatically run it over a pattern or list of targets.
You can test your scripts interactively in headless mode (that is, without a browser). You can also run batch either with a browser, or headless.
Knowledge You Should Have¶
You should have a general understanding of HTML and CSS structure and form. You don’t need to know much, but you should be able to understand and recognize what you are looking at when looking at small portions of web page source, and have an understanding of what type of thing you are trying to extract, i.e. path, attribute, or text.
You will need some basic understanding of XPATH syntax and CSS Selectors as you will be using these to describe what you are looking for. When manually highlighting something in your browser, [S]crape will return an XPATH. Often a CSS selector is both shorter and more accurately selective. [S]crape allows you to view context near your selection. This makes it easy to pick a different form of selector and test it before saving it to your script.
Shell¶
In interactive use, [S]crape is similar to a typical command shell,
such as sh or bash, or cmd on Windows.
In command interpreters, there are typically built-in commands
and a way to execute external commands.
Shells also provide variables, and some sort of program control.
[S]crape has a rich set of built-in commands, and allows calling external commands through your system’s shell. You can also add built-in commands by writing extensions to [S]crape in Python (plugins).
Since [S]crape outputs tables [2],
variable names are like table column names.
This means every variable in [S]crape is a list (you can think of them as arrays),
and every table an associative array of variables.
In fact, you can save the result of your [S]crape as either csv, json or yaml.
There are other important kinds of variables in [S]crape.
| vars: | Output variables are the normal variables, and are used to specify output table column names. |
|---|---|
| local: | Local variables are similar to output variables, only they are omitted from tables. These are used for intermediate results. Local variables have scope per output table. |
| global: | These variables persist across output table changes. |
[S]crape is least like shells in that there is no familiar loop control. This simplifies traversing an HTML tree and extracting data. Instead of looping, you traverse to locations in the XPATH tree of the input file. We refer to selected (current) XPATH locations as nodes. Typical [S]crape operation involves traversing a document’s tree, extracting selected content from those nodes, and repeating. In place of program control, you control which nodes you search from. Multiple nodes can be active (for example all the list items of some part of the document), so scripts tend to be rather short. Some general control mechanisms [S]crape provides are:
| root: | Normally, navigation through the document is incremental.
This sets the root of the tree to the starting <html> tag.
When the root of the document tree is set, it’s children are the
active children, so in this case, normally <head> and <body>
tags will be current starting nodes. |
|---|---|
| body: | This resets the root node to the <body> tag. |
| grab: | [S]crape opens a browser when it starts, and communicates with it.
grab gets a highlited region from your browser,
giving you an xpath to it. |
A majority of [S]crape commands involve selecting a node using an xpath selector, a css-selector, or a combination of path and text search. The remaining commands deal with interactive use (history, view variables, run scripts, save or load scripts), and outputing results (tables).
Footnotes
| [1] | [S]crape uses a Selenium Client Driver to run your browser. At this time [S]crape only supports Firefox. |
| [2] | [S]crape was initially designed to output CSV, but this is a bit too restricting. For one thing, to change the view of the data (the order of way the data is populated into columns, the number and contents of tables) one would need to re-scrape the source. This is why you have a choice of saving variables as JSON or YAML also. Then, you could rebuild, re-shape your tables from your saved data source. |
Installation¶
requires the following:
Installing [S]crape will install or upgrade the following python libraries:
- argparse
- lxml
- cssselect
- PyYAML
- selenium
- [S]crape‘s version of envoy
- a [S]crape plugin library
Installation requires you have a compiler on your machine.
For Linux systems, this should already be the case.
For Macintosh OS/X systems, downloaded Xcode free from the Mac App Store (also install the command line tools). Alternatively, you may be able to install just the command-line tools (see https://github.com/kennethreitz/osx-gcc-installer - we have not tried this).
For Windows platforms, this is not a straightforward process.
See the MS Windows section at http://lxml.de/installation.html.
Installing ¶
I suggest you use python’s virtualenv, particularly your first
time with [S]crape
(see virtualenv).
This will ensure you have an isolated, clean python install of [S]crape to start. Once you have this working, you may consider installing this your system’s python site-libraries.
To properly use virtualenv, you’ll need pip.
Ensure you have pip installed:
$ which pip
If you don’t have pip installed, then install it:
$ easy_install pip
If you do have pip, be sure it’s up-to-date:
$ pip install --upgrade pip
Todo
Have yet to debug the scrape.gz install file (installation does not mirror setup.py).
Now, install the current version of [S]crape. Currently, you must do this from sources. Clone a copy of [S]crape and run setup.py:
$ hg clone ssh://hg@bitbucket.org/yarko/scrape
$ cd scrape
$ python setup.py install
Footnotes
| [1] | Firefox is the only browser officially supported for [S]crape. As an alternative, you may try a current version of Chrome, but note that you will need to download a chrome-webdriver. For some combinations of versions of Chrome, chrome-webdriver and selenium, timeouts didn’t properly work. For some medical journal sites with continuous stream advertising, Chrome would not respond (would never return when called from scrape). |
Tutorials¶
Select the tutorials which are appropriate for what you want to do.
To start, I recommend you follow the installation checkout and tutorial on this page, followed by the example PyCon project. These brief tutorials will introduce the concepts and strategies for using scrape, as well as give an overview of some of the most useful commands.
![[S]crape tutorials](_images/graphviz-1a584fa556b22fe87b08ee9e8ad17a12546f2160.png)
Introduction to [S]crape¶
Getting Started¶
Overview¶
Introduction¶
We’ll start by making sure you have [S]crape installed.
To start, I’ll assume you have installed [S]crape in a virtualenv.
Activate the virtual environment where you have installed [S]crape,
and run scrape:
$ scrape http://scrape.readthedocs.org
This should start [S]crape and open its documentation in Firefox [2].
your Firefox should have “WebDriver” displayed in the lower-right;
- this indicates that this Firefox is being controlled from [S]crape.
you should see a log of the plugins registered (scrape comes distributed with one -
affiliations);
- if you don’t see
scrape: INFO - ...registering plugins:in your [S]crape shell, then likely something is incomplete in your installation. You can continue with the exercises, but you will need to install plugins when you need them.
At the [S]crape >>> prompt type the following:
[S]crape >>> help
[S]crape gives you access to an HTML file or web page. It does this by parsing your web page into a tree of HTML nodes. You then traverse the tree of nodes, scraping the information you want from a selected nodes.
[S]crape starts by setting the root node of your HTML page [1]
to the <html> node.
Let’s show the contents of the current node:
[S]crape >>> show node
You should see the source for the two subnodes (children) of the <html> tag,
the <head> and <body> tags of the [S]crape document page.
This is the content of the document, rooted at html [3].
Just to confirm, lets count the current number of selected nodes:
[S]crape >>> nodes
When you will be looking at larger, more verbose selections, it can also be helpful to review just the tags of the selected nodes:
[S]crape >>> tags
The general starategy for using [S]crape is:
- select a scrape target (a web page);
- declare a table name (which will hold a set of variables);
- navigate the web page’s tree;
- declare a variable to collect information;
- capture the desired information;
- repeat as desired;
- save a table to a file;
- lastly, save your interactive commands to a script to run later.
When looking a [S]crape target,
it’s useful to also open the source view.
When you are starting with a new page,
you can easily search the various
tags and attributes of the html elements.
Do that now - right-click on your webpage, and select
View Page Source.
For this tutorial, we’ll save the headers to develop an outline for this page.
The outline we’d like to make consists of the headers, Contents:, Alternatives,
and so forth. In your source window, search for Contents:.
Contents: is in an <h2> tag, as is Alternatives; this looks like a reasonable
target for our scrape.
[S]crape provides a simplified interface to the libxml2 library,
so that most of the information you will find about xpath selectors
and cssselectors will work as you expect.
[S]crape also combines, extends and adds other commands
for interactive use.
For example, find_by_text will search
nodes selected by an xpath expression for a string.
You might like view http://www.w3schools.com/xpath/xpath_syntax.asp for reference during this tutorial.
Let’s find the subheadings on our target page to see if this will give us the page outline we’d like:
[S]crape >>> findall .//h2
[S]crape >>> show node
This should find all the <H2> nodes under the current node.
More than one node is found - show displays all of the currently selected nodes.
There are four active nodes, as verified by:
[S]crape >>> nodes
The text of these nodes seems like it would serve nicely as an outline, so lets capture those.
First, declare a table name and a variable to collect output (if you don’t declare a table name,
the default is scrape_table).
Setting var my_name will select a variable to collect data.
The variable does not need to exist (it will be created).
If you change tables before you’ve saved their output,
they are stored so you can later add to their variables (and output).
[S]crape >>> table outline
[S]crape >>> var topics
When you save the output from this table, it will be saved in a file outline.csv.
You can also save the output as json or yaml.
Once you save a table, its values are emptied.
So far, this table has one column - one variable.
To see what the various variables of a table currently have,
we issue the show out command
to show pending output (the current table’s contents).
[S]crape variables are lists of values.
Varible names are shown with a colon (my_var:),
and their values are shown preceded by a ‘-‘.
The text command will collect text contents of the currently
selected HTML nodes into the current variable.
[S]crape >>> show out
[S]crape >>> text
[S]crape >>> show out
There was no output pending prior to the text command.
If you wanted to save this now, the table command (with no argument)
will output the current table to a csv file with the same name
(if one already exsits, it will not be overwritten;
the name will be numerically extended).
If you want to save your script for later, look at your history. Only scrape commands which act on pages are saved in history. You can choose which parts of history you save to a script file.
[S]crape >>> history
[S]crape >>> help save
If you’d like, save your script now. You can edit saved [S]crape scripts with a text editor. You can add comments, which begin with ‘#’ and extend to the end of the line.
There is an alternate form for selecting tables and variables, which may help the commands in your script (and what they apply to) stand out. If you’d like, in place of:
table outline
var topics
you can equivalently write:
[ outline ]
< topics >
To exit [S]crape, see help EOF.
After our brief interactive session with [S]crape, here’s what our script looks like:
##
# [S]crape script to get outline of a page
#
# - gets the text of <h2> headings;
#
[ outline ]
< topics >
findall .//h2
text
table # save outline.csv
Summary¶
In this introductory tutorial, we’ve
- shown one way to select nodes;
- defined tables and variables;
- saved selected content;
- saved a [S]crape script;
Please continue with the next tutorial.
Happy ing!
Footnotes
| [1] | You can easily set to the root of the document at any time to either the entire document,
or the body - see help doc and help body. |
| [2] | Be sure you’ve installed [S]crape and a current Firefox browser. |
| [3] | Note that when you look at an empty page (about:blank),
scrape will create a minimal parse tree for you (<html><head/><body/></head>). |
Developing a Project¶
This is the second tutorial in a series.
From the introductory tutorial, we saw how
to select a destination (file or URL).
Initially, it’s also beneficial to
view the destination’s source along with the
browser window.
You can either search for what interests you
in the source window, or use inspect element to
get to the item that interests you.
Once you’ve quickly found the item of interest, you can start trying various tree traversal commands to get to related items in [S]crape, view the nodes found, and save some part of their content in [S]crape variables for output. You can also save your script activity into a script, which you can edit and run later in [S]crape.
In this tutorial we’ll see how to backtrack and make corrections. We’ll also see how the various [S]crape commands behave when applied to multiple nodes.
Note
A word about [S]crape ing public sites:
Be a Good Citizen!
- avoid repeatedly hitting a site, and loading its servers;
- always check for copyright, and observe fair use doctrines.
PyCon Volunteer Reporting¶
Here’s our project: the US PyCon 2013 Conference is coming up. PyCon is a community conference and depends heavily on voluneers. We want to track how many volunteers we still need for session staff [1].
The conference site lists the sessions and staff on http://us.pycon.org/2013/schedule/sessions. Since this will likely change dynamically, we’ll use a snapshot version we saved, just as you would when first developing a script (in order to spare repeatedly hitting a site’s servers). Having a static copy will also make it easier to follow along with the tutorial (also, after the conference, there will be no unfulfilled needs, so the web data won’t be as interesting):
- download
tutorial2.zip.
Getting Oriented with [S]crape Commands¶
Let’s review what we’ve learned so far.
When you open [S]crape with a URL, [S]crape opens
the url in a browser and parses it into a tree of nodes
held in scrape.
These nodes are what you navigate.
Using xpath and cssselect you select nodes and
extract data.
The ability to inspect aspects during the process is useful, as well as being able to run scripts in batch.
In this tutorial we’ll introduce some of the rhyme and reason behind [S]crape. Since [S]crape has over 60 commands, let’s start by describing some structure around the commands (we will only introduce some of them in this tutorial).
![[S]crape Context](_images/graphviz-5b80bd82137457c5d465f221bd1045db439506e5.png)
[S]crape commands affect each of these areas. Most of the action happens in the hub - in [S]crape itself. The type of commands in [S]crape are:
- navigation
- content extraction (capturing)
- interaction
- settings
- variables
A Starting Strategy¶
The first time you open a target URL it can be useful
to open the page’s source from the browser (I have them
side-by-side at first).
right-click in the browser page:
For smaller pages, it can be useful to search in the source
for what interested you in the browser.
For larger pages, it can sometimes be easier to simply
highlight what interests you in the web page, and use
the [S]crape grab command to give you a small context.
From there, it can be easier to search for the larger
context in the source window, so you can get a good
view of the context around your interest.
Let’s do that now.
Unzip the tutorial file (I’ve replaced the >1M in images with
a single pixel gif to keep things manageable).
You should have a file sessions.html and a directory
sessions_files.
Assuming you’ve unzipped in the current directory
run scrape:
$ scrape sessions.html
To orient ourselves, use a few of the interaction commands from the Introductory Tutorial:
[S]crape >>> nodes
2
[S]crape >>> tags
['head', 'body']
[S]crape >>>
In this case, we are not concerned with any of the meta-data
which might be in the <head/>:
[S]crape >>> body
[S]crape >>> nodes
7
[S]crape >>> tags
['header', 'div', 'script', 'script', 'script', 'script', 'div']
[S]crape >>>
Looking at our browser window, the sessions are named and listed as visual blocks. Here are the parts interesting for our task:
Scrolling to the bottom of the browser page, we see there are 42 sessions.
We can see that each session has a Session Chair and a Session Runner.
If no one has signed up, the page shows: No volunteers signed up.
We need a total of 84 volunteers. We’ll need to gather information
after the session name (e.g. Session #1).
Unfortunately, there’s a lot of HTML code for headers, sponsors,
and so forth - but let’s go to our browser’s source window and search for Sessions.
It looks like our info is all contained in an HTML list.
Let’s just start by seeing what happens when we try to get the list of sessions. If we try findclass:
[S]crape >>> findclass unstyled
[S]crape >>> nodes
43
[S]crape >>>
It looks like we might have gotten the 42 session (their content looks to be held
in <ul class="unstyled"> lists), and the outermost list holding them.
You can look at what was selected with show node, but it’s a little easier to
digest at this point in the browser-source window.
This is close to what we wanted, but not quite.
Adjusting Course¶
If you use further [S]crape navigation commands (such as findclass), they will act from each of the currently selected nodes. We’re not where we want to be, so let’s back up:
[S]crape >>> body
Some smaller ways you can back up in the tree:
- doc, or root (aliases)
- getprevious
- getparent
See the help for these, and experiment with them.
Now, let’s try a couple of other commands to see if you can get to the 42 nodes of interest (look for hints in the browser-source view). Here are a few examples (I’ll omit the output, so be sure to follow along at your computer):
[S]crape >>> flindclass unstyled
[S]crape >>> nodes
[S]crape >>> tags
[S]crape >>> body
[S]crape >>> help cssselect
[S]crape >>> cssselect div.box-content ul.unstyled
[S]crape >>> nodes
[S]crape >>> tags
There are a couple of ways to get to what we want (you may find others).
Cssselectors are easy to write and powerful.
Xpath expressions are explicit and functional (if you learn xpath expressions,
you can take advantage of that knowledge for navigating XML documents also).
I find that either cssselect h1+ul.unstyled or find .//ul[@class='unstyled']
work.
The css expression says:
get all the elementsulwhich immediately follow anh1, and which have classunstyled.
The xpath expression says:
get the next (single)ulnode with classunstyled.
The .//tag_name form says
look anywhere (any depth) under the current node.
I prefer the xpath expression - for this case, it seems more suitable, closer to what we intend.
[S]crape >>> body
[S]crape >>> find .//ul[@class="unstyled"]
[S]crape >>> nodes
[S]crape >>> show node
This looks like the spot we were interested in, in the browser-source.
Saving Output¶
So that we can have context, let’s collect the session name.
Let’s also scrape the text of the first ul under that - the session volunteers.
I want to have 42 names, and 42 pieces of volunteer information.
Thus we can determine which sessions have needs.
The first ul under each session name will do this for us.
First, lets try to select the sessions. From the last show command, we can
see Session #42.
[S]crape >>> findall ./li
[S]crape >>> nodes
[S]crape >>> show
Note that findall has a single '/' -
this will find only direct children of our current ul node.
Now lets get our session names:
[S]crape >>> find ./a
[S]crape >>> nodes
[S]crape >>> show
We use find (not findall) because we only want the first a tag under each of our 42 nodes.
This time, the show command is a joy to look at - it’s clear that we have the session names, that our 42 nodes are indeed exactly what we want. The text of these nodes contain the session names we want. We’re ready to setup some variables:
[S]crape >>> [sessions_table]
[S]crape >>> <session>
[S]crape >>> text
[S]crape >>> show out
We have our 42 session names waiting to be output. But still we need to add information about the volunteer status of each.
Thankfully, we have the browser-source window to refer to.
We can see that after the <a> containing our session names we want
the ul nodes which are the first children of the div tag following li.
The getnext command will get the next sibling node (the div we want).
From there we will get the ul directly under:
[S]crape >>> getnext
[S]crape >>> nodes # confirm
[S]crape >>> show
[S]crape >>> find ./ul
[S]crape >>> nodes # still looks good
Where text will get the text inside the tag (up to the next child tag), text_content will get all the text inside a tag, even that inside other enclosed nodes. We’re ready to save the status of the volunteers - we’ll put this in a volunteer variable.
[S]crape >>> <volunteer>
[S]crape >>> text_content
[S]crape >>> show out
There is a good deal of white space, but we’ll easily deal with that outside of scrape. I think the form of show out (yaml) would be easy to read into a python script which will do the counting.
[S]crape >>> yaml sessions.yaml
You could have also saved this as either json or csv (the latter using the table command). Either json or yaml is convenient for loading into python data structures. I chose yaml because it is easy on the eyes when viewing the scraped data file.
Running Another Day¶
We’ll need to run this script quite often to keep the current volunteer needs up to date, so we’ll need to save our script.
Have a look at your history:
[S]crape >>> history
Notice that history shows your navigation commands, but not your interactive insepction commands. Scripts are saved from this command history, so inspection commands are not stored there.
You could edit your script file (comments start with '#'),
and eliminate any false starts, and test the edited result, or
you could select which part of your history to save, and go from there.
You decide:
[S]crape >>> help save
[S]crape >>> save sessions.scrape
Before you exit [S]crape, edit your file, and test it by running it against the current page (I use the gvim editor; you should use your favorite):
[S]crape >>> clear volunteers session # clear variables
[S]crape >>> show out # should now have no output pending
[S]crape >>> load sessions.scrape
[S]crape >>> show out
Loading a script runs it against the current document tree.
You can run your script in headless mode:
$ scrape -H -s sessions.scrape http://us.pycon.org/2013/schedule/sessions
I leave it to you to develop a script to count and report on volunteer needs, based on sessions.yaml. Mine was under 12 lines of python. Whatever you use for postprocessing, you can also run it from your sessions.scrape by adding something like this to the bottom of your script:
# after saving your yaml / json / csv file:
!python my_script.py sessions.yaml
Summary¶
After this exercise, your script should look similar:
## Count volunteer signups for PyCon Sessions
#
# open http://us.pycon.org/2013/schedule/sessions/
#
body
# I save to a different name than this table, which would be default;
[s_table]
find .//ul[@class='unstyled']
findall ./li
find ./a
<session> # column1: the session name
text
getnext
find ./ul
<txt> # column2: who's signed up to staff the session;
text_content
yaml sessions
!python session_volunteer_counter.py sessions.yaml
Let’s look at which commands in [S]crape we used:
- navigation:
- body
- cssselect
- find
- findall
- findclass
- getnext
- capturing:
- text
- text_content
- interaction:
- help
- history
- nodes
- show
- show out
- tags
- settings:
- headless (-H)
- variables:
- clear
- “[...]”, or table
- “<...>”, or var
- output:
- yaml
- scripts:
- load
- save
- shell:
- ”!”, or shell
Happy ing!
Footnotes
| [1] | Session staff consist of chairs and runners. Chairs introduce speakers, manage questions and keep track of time. Runners get speakers to their talk on time and ensure they have everything they need. A typical session consists of 3 talks. Sessions run simultaneously in multiple rooms. |
Introduction to [S]crape¶
The “Getting Started” Tutorial will help you confirm your installation, and introduce the basic
Developing a Project¶
In this Intermediate Tutorial, we’ll look at using
Copyright Notice¶
Copyright (c) 2013, The University of Chicago. All rights reserved.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
- Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
- Neither the name of The University of Chicago nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Installation¶
Software and browser requirements.
Alternatives¶
Examples of just some of the alternatives to [S]crape include:
You can also look through the notable tools section on Wikipedia.
Many of these are either not interactive, or are programmers libraries or toolkits. [S]crape is an interactive script development tool which, with a modicum of knowledge, is both powerful and simple. [S]crape scripts use [S]crape commands, shell commands, and commands provided by extensions.
Copyright Notice¶
This license applies to the program,
Work in Progress¶
This document is currently a work-in-progress. Here are a list of known items left to do:
Todo
Have yet to debug the scrape.gz install file (installation does not mirror setup.py).
(The original entry is located in /home/docs/checkouts/readthedocs.org/user_builds/scrape/checkouts/latest/source/installation.rst, line 92.)