Introduction to web scraping

I have been asked to host a tutorial on web scraping for my research group. I figured I’d blog it so that it can be a reference in the future as well.

# GitHub

I’m not going to extol the virtues of GitHub, but rather assume that everyone is using it (or GitLab etc.).
In your online account: create the repository, set up a .gitignore, and then clone it into the folder (in the parent folder of your project - i.e. if you want your scraper directory to be within research, be in the research folder when you clone.

If you are unfamiliar with Git, check out the PowerPoint downloadable from my earlier blog.

# Directory structure

|- <your project name>/
|- .git (should be here already)
|- .gitignore (should be here already)
|- _scratch_/
|- data/
|- src/ (where the code will go)


# Code editor

I suggest working in Atom (because of it’s Git integration). But Sublime or others are also good.

# Virtual Environment

My example will be for scraping in Python, so, as with any Python coding project, we’ll be working in a virtual environment.
Create one in the highest level of your project’s directory. That is, in the code terminal you’re using (OS dependent)

1. (if you haven’t already you’ll need to install the ability to create virtual environments: pip install virtualenv)
2. cd to your project’s directory.
3. python3 -m virtualenv venv or python -m virtualenv venv

This will add venv directory into your project directory. Remember to add venv/ to the .gitignore file (this is easy in most code editors).
Now Git commit.

# Prepare the virtual environment

There are a couple of packages that we will use to perform the web scraping, so install them into your virtual environment. The first step to doing this is to activate the virtual environment. To do this, type into your terminal (once you’re in the project directory):
Mac/Linux: source venv/bin/activate
Windows: venv\Scripts\activate
The result should be a (venv) showing on the left on your terminal active line.

We can now install packages into the virtual environment without administrator privileges.
Go ahead and install the requests, beautifulsoup, and schedule packages with: pip install requests bs4 schedule

# E.g. Power outage scraping

Let’s scrape the electricity outages for the counties served by BGE.
Open the BGE site that has these outages: https://outagemap.bge.com/

The irritating thing with many of these websites is how often they change, so I’ll provide a generally description so (a) it’s extendable to other sites, and (b) this tutorial isn’t made completely redundant when they next update their website.

The aim here is to regularly record the number of customers in each county without power. So the first step is to find that information on the site. In this case, this is available on the left hand side of the screen, click summary and then Outages by County. The result is a tabular popup that lists the information.

The next step is to figure out where that information is stored by the website. Most websites are comprised of data and code that formats and displays that data. We need to figure out where and how that data is stored. To look under-the-hood of the website we need to Inspect it (Chrome Windows: Crtl+Shift+I; Safari Mac: Command+Option+i - or Google the command for your browser and OS.) Once the inspector is open, refresh the page.

Now we need to find the data within this list of files. Often, and in this case, it’s a .js file. For data such as this, it comes with a timestamp. Because of variation between websites, you need to search this list and look at the result in the box. Practice, and getting a feel for what you’re looking for, makes this process faster.

Now that you’ve found the information in the website that you want, you need to figure out how to access it. Right click on the name (in this case report.js?timestamp=1551... etc) and click Open in new tab. This is the text that we’ll be downloading. (I can’t stress enough, that websites are different and some will simply not let you do this. In those cases it does not mean that they cannot be scraped - but you need to dig more to figure out how. The ones that provide data in these json/js formats are ideal, but there are other formats that can be scraped.) Now copy the URL: e.g. https://s3.amazonaws.com/outagemap.bge.com/data/interval_generation_data/2019_02_28_12_26_08/report.js?timestamp=1551374668531 (you can actually drop the info after the timestamp).

For this website, there’s one more thing you need from the website and then we can turn to the code. That’s where the datetime stamp is stored. We know what the URL is, but the datetime stamp is a variable and that comes from the website. Go back to the inspector and find where that datetime information is. In this case it is in metadata.xml

So, following the same steps, we can get the URL as: https://s3.amazonaws.com/outagemap.bge.com/data/alerts/metadata.xml which is kindly datetime independent.

Now over to the code.

I recommend the structure of this code be along the lines of:

1. wrapper function that runs every x minutes.
2. function that scrapes and returns the data
3. function that writes to a text file or, preferably, a database.

## Scheduler

This is the basic code for a scheduler.

## Scraper code

First off, let’s see if we can scrape that URL you found earlier. Try: content = requests.get(url).content
Now print content and see that it prints what you saw on the browser.

So here is an example of a scraping function.

We first get the datetime stamp, then update the data’s URL so that we can download that. Finally we decode that data into a dictionary so we can access the data next.

Note that the last line calls a function write_data, so we need to figure this out!

## Storing data

An essential element of this is how you’re storing storing this data. I thoroughly recommend InfluxDB as it is a temporal database with great documentation. But that’s for another time. So for now, we’ll store it in a text file. But what is the structure of the data? Every data set is different and considering its future use is critical to how it is stored.

It would be easy to store the data with each row being a timestamp. But I think there are better ways. Instead, let’s store each county outage as a row. That means that each timestamp will be repeated. Doing this means we can quickly filter by county, time, number of customers out etc.

That is, the data structure I suggest we use is:

datetime | county_name | custs_out | custs_total | provider


Provider is optional here and would come in handy if you were to have a single dataset for multiple utilities.

So, back to the code.

Look at the format of the data in data_dict.
This is a dictionary in Python. This becomes very useful as we can access the different components. Print data_dict and look at the keys. The county data is actually a list of dictionaries stored here: data_dict['file_data']['curr_custs_aff']['areas'][0]['areas']
So what we’ll do is loop through these and write them to rows in the textfile/database.

# Putting it all together

And this should be all you need. Save this as something like elect_scraper.py and from the terminal, in the upper level of this directory, you can run it with python src/elect_scraper.py and it’ll save data into the data/ folder.

#### Previous

Neglecting future urban development and changing risks can lead to maladaptation to natural hazards

Last week Hurricane Florence caused, and continues to cause, major damage along the East Coast of the USA. Following Florence and a year of record breaking heat waves and fires now is a good time to consider adapting to climate change and mitigating natural hazards. Are strategies we’re taking really going to work? Or is it possible that the engineering approaches taken to protect us are actually just making us more vulnerable?

#### You May Also Enjoy

Python vs R: Random Forests

Time for the speed and accuracy comparison

Better Coding Practice

Have you ever had to revisit a coding project? If you're like me, there's a sense of trepidation. Will I remember what I was doing? Will it still work? Here are some techniques that may make you a happier and more efficient researcher.

Watercolor (aka Visually Weighted) regressions in Python

Scatter plots with coloured bands for the distribution.

Documenting literature reviews