Tag: data science

Nov 01 2013

Book review: Tableau 8 – the official guide by George Peck

tableau 8 guideThis review was first published at ScraperWiki.

A while back I reviewed Larry Keller’s book The Tableau 8.0 Training Manual, at the same time I ordered George Peck’s book Tableau 8: the official guide. It’s just arrived. The book comes with a DVD containing bonus videos featuring George Peck’s warm, friendly tones and example workbooks. I must admit to being mildly nonplussed at receiving optical media, my ultrabook lacking an appropriate drive, but I dug out the USB optical drive to load them up. Providing an online link would have allowed the inclusion of up to date material, perhaps covering the version 8.1 announcement.

Tableau is a data visualisation application, aimed at the business intelligence area and optimised to look at database shaped data. I’m using Tableau on a lot of the larger datasets we get at ScraperWiki for sense checking and analysis.

Colleagues have noted that analysis in Tableau looks like me randomly poking buttons in the interface. From Peck’s book I learn that the order in which I carry out random clicking is important since Tableau will make a decision on what you want to see based both on what you have clicked and also its current state.

To my mind the heavy reliance on the graphical interface is one of the drawbacks of Tableau, but clearly, to business intelligence users and journalists, it’s the program’s greatest benefit. It’s a drawback because capturing what you’ve done in a GUI is tricky. Some of the scripting/version control capability is retained since most Tableau files are in plain XML format with which a little fiddling is tacitly approved by Tableau – although you won’t find such info in The Official Guide. I’ve been experimenting with using git source control on workbook files, and it works.

If you’re interested in these more advanced techniques then the Tableau Knowledgebase is worth a look. See this article, for example, on making a custom colour palette. I also like the Information Lab blog, 5 things I wish I knew about Tableau when I started and UK Area Polygon Mapping in Tableau. The second post covers one of the bug-bears for non-US users of Tableau: the mapping functionality is quite US-centric.

Peck covers most of the functionality of Tableau, including data connections, making visualisations, a detailed look at mapping, dashboards and so forth. I was somewhat bemused to see the scatter plot described as “esoteric”. This highlights the background of those typically using Tableau: business people not physical scientists, and not necessarily business people who understand database query languages. Hence the heavy reliance on a graphical user interface.

I particularly liked the chapters on data connections which also described the various set, group and combine operations. Finally I understand the difference between data blending and data joining: joining is done at source between tables on the same database whilst blending is done on data from different sources by Tableau, after it has been loaded. The end result is not really different.

I now understand the point of table calculations – they’re for the times when you can’t work out your SQL query. Peck uses different language from Tableau in describing table calculations. He uses “direction” to refer to the order in which cells are processed and “scope” to refer to the groups over which cell calculations are performed. Tableau uses the terms “addressing” and “partitioning” for these two concepts, respectively.

Peck isn’t very explicit about the deep connections between SQL and Tableau but makes sufficient mention of the underlying processes to be useful.

It was nice to see a brief, clear description of the options for publishing Tableau workbooks. Public is handy and free if you want to publish to all. Tableau Online presents a useful halfway house for internal publication whilst Tableau Server gives full flexibility in scheduling updates to data and publishing to a range of audiences with different permission levels. This is something we’re interested in at ScraperWiki.

The book ends with an Appendix of functions available for field calculations.

In some ways Larry Keller and George Peck’s books complement each other, Larry’s book (which I reviewed here) contains the examples that George’s lacks and George’s some of the more in depth discussion missing from Larry’s book.

Overall: a nicely produced book with high production values, good but not encyclopedic coverage.

Oct 29 2013

Git!

logo@2x

This post was first published at ScraperWiki.

As software company, use of some sort of software source control system is inevitable, indeed our CEO wrote TortoiseCVS – a file system overlay for the early CVS source control system. For those uninitiated in the joys of software engineering: source control is a system for recording the history of file revisions allowing programmers to edit their code, safe in the knowledge that they can always revert to a previous good state of code if it all goes horribly wrong. We use Git for source control, hosted either on Github or on Bitbucket. The differing needs of our platform and data services teams fit the payment plans of the two different sites.

Git is a distributed source control system created by Linus Torvalds, to support the development of Linux. Git is an incredibly flexible system which allows you to do pretty much anything. But what should you do? What should be your strategy for collective code development? It’s easy to look up a particular command to do a particular thing, but less is written on how you should string your git commands together. Here we hope to address this lack.

We use the “No Switch Yard” methodology, this involves creating branches from the master branch on which to develop new features and regularly rebasing against the master branch so that when the time comes the feature branch can be merged into the master branch via a pull request with little fuss. We should not be producing a byzantine system by branching feature branches from other feature branches. The aim of “No Switch Yard” is to make the history as simple as possible and make merging branches back onto master as easy as possible.

How do I start?

Assuming that you already have some code in a repository, create a local clone of that repository:

git clone [email protected]:scraperwiki/myproject.git

Create a branch:

git checkout -b my-new-stuff

Start coding…adding files and committing changes as you go:

git add -u
git commit -m "everything is great"

The -u switch to git add simply checks in all the tracked, uncommitted files. Depending on your levels of paranoia you can push your branch back to the remote repository:

git push

How do I understand what’s going on?

For me the key revelation for workflow was to be able to find out my current state and feel pleasure when it was good! To do this, fetch any changes that may have been made on your repository:

git fetch

and then run:

git log --oneline --graph --decorate --all

To see an ASCII art history diagram for your repository. What you are looking for here is a relatively simple branching structure without too many parallel tracks and with the tips of each branch lined up between your local and the remote copy.
You can make an alias to simplify this inspection:

git config --global alias.lg 'log --oneline --graph --decorate'

Then you can just do:

git lg --all

I know someone else has pushed to the master branch from which I branched – what should I do?

If stuff is going on on your master branch, perhaps because your changes are taking a while to complete, you should rebase. You should also do this just before submitting a pull request to merge your work with the master branch.

git rebase -i

Allows you to rebase interactively, this means you can combine multiple commits into a single larger commit. You might want to do this if you made lots of little commits whilst achieving a single goal. Rebasing brings you up to date with another branch, without actually merging your changes into that branch.

I’m done, how do I give my colleagues the opportunity to work on my great new features?

You need to rebase against the remote branch onto which you wish to merge your code and then submit a pull request for your changes. You can submit a pull request from the web interface at Github or Bitbucket. Or you can use a command line tool such as hub.  The idea of using a pull request is that it makes your changes visible to your colleagues, and keeps a clear record of those changes. If you’ve been rebasing regularly you should be able to merge your code automatically.

An important principle here is “ownership”, in social terms you own your local branch on which you are developing a feature, so you can do what you like with it. The master branch from which you started work is in collective ownership so you should only merge changes onto it with the permission of your colleagues and ideally you want others to look at your changes and approve the pull themselves.

I started doing some fiddling around with my code and now I realise it’s serious and I want to put it on a branch, what did I do?

You need to stash your code, using:

git stash

Then create a branch, as described above, and then retrieve the contents of the stash:

git stash pop

That’s how we use git – what do you do?

Aug 27 2013

Book review: The Tableau 8.0 Training Manual – From clutter to clarity by Larry Keller

Tableau 8.0 Training Manual

This review was first published at ScraperWiki.

My unstoppable reading continues, this time I’ve polished off The Tableau 8.0 Training Manual: From Clutter to Clarity by Larry Keller. This post is part review of the book, and part review of Tableau.

Tableau is a data visualisation application which grew out of academic research on visualising databases. I’ve used Tableau Public a little bit in the past. Tableau Public is a free version of Tableau which only supports public data i.e. great for playing around with but not so good for commercial work. Tableau is an important tool in the business intelligence area, useful for getting a quick view on data in databases and something our customers use, so we are interested in providing Tableau integration with the ScraperWiki platform.

The user interface for Tableau is moderately complex, hence my desire for a little directed learning. Tableau has a pretty good set of training videos and help pages online but this is no good to me since I do a lot of my reading on my commute where internet connectivity is poor.

Tableau is rather different to the plotting packages I’m used to using for data analysis. This comes back to the types of data I’m familiar with. As someone with a background in physical sciences I’m used to dealing with data which comprises a couple of vectors of continuous variables. So for example, if I’m doing spectroscopy then I’d expect to get a pair of vectors: the wavelength of light and the measured intensity of light at those wavelengths. Things do get more complicated than this, if I were doing a scattering experiment then I’d get an intensity and a direction (or possibly two directions). However, fundamentally the data is relatively straightforward.

Tableau is crafted to look at mixtures of continuous and categorical data, stored in a database table. Tableau comes with some sample datasets, one of which is sales data from superstores across the US which illustrates this well. This dataset has line entries of individual items sold with sale location data, product and customer (categorical) data alongside cost and profit (continuous) data. It is possible to plot continuous data but it isn’t Tableau’s forte.

Tableau expects data to be delivered in “clean” form, where “clean” means that spreadsheets and separated value files must be presented with a single header line with columns which contain data all of the same type. Tableau will also connect directly to a variety of databases. Tableau uses the Microsoft JET database engine to store it’s data, I know this because for some data unsightly wrangling is required to load data in the correct format. Once data is loaded Tableau’s performance is pretty good, I’ve been playing with the MOT data which is 50,000,000 or so lines, which for the range of operations I tried turned out to be fairly painless.

Turning to Larry Keller’s book, The Tableau 8.0 Training Manual: From Clutter to Clarity, this is one of few books currently available relating to the 8.0 release of Tableau. As described in the title it is a training manual, based on the courses that Larry delivers. The presentation is straightforward and unrelenting; during the course of the book you build 8 Tableau workbooks, in small, explicitly described steps. I worked through these in about 12 hours of screen time, and at the end of it I feel rather more comfortable using Tableau, if not expert. The coverage of Tableau’s functionality seems to be good, if not deep – that’s to say that as I look around the Tableau interface now I can at least say “I remember being here before”.

Some of the Tableau functionality I find a bit odd, for example I’m used to seeing box plots generated using R, or similar statistical package. From Clutter to Clarity shows how to make “box plots” but they look completely different. Similarly, I have a view as to what a heat map looks like and the Tableau implementation is not what I was expecting.

Personally I would have preferred a bit more explanation as to what I was doing. In common with Andy Kirk’s book on data visualisation I can see this book supplementing the presented course nicely, with the trainer providing some of the “why”. The book comes with some sample workbooks, available on request – apparently directly from the author whose email response time is uncannily quick.

Aug 23 2013

Making a ScraperWiki view with R

 

This post was first published at ScraperWiki.

In a recent post I showed how to use the ScraperWiki Twitter Search Tool to capture tweets for analysis. I demonstrated this using a search on the #InspiringWomen hashtag, using Tableau to generate a visualisation.

Here I’m going to show a tool made using the R statistical programming language which can be used to view any Twitter Search dataset. R is very widely used in both academia and industry to carry out statistical analysis. It is open source and has a large community of users who are actively developing new libraries with new functionality.

Although this viewer is a trivial example, it can be used as a template for any other R-based viewer. To break the suspense this is what the output of the tool looks like:

R-view

The tool updates when the underlying data is updated, the Twitter Search tool checks for new tweets on an hourly basis. The tool shows the number of tweets found and a histogram of the times at which they were tweeted. To limit the time taken to generate a view the number of tweets is limited to 40,000. The histogram uses bins of one minute, so the vertical axis shows tweets per minute.

The code can all be found in this BitBucket repository.

The viewer is based on the knitr package for R, which generates reports in specified formats (HTML, PDF etc) from a source template file which contains R commands which are executed to generate content. In this case we use Rhtml, rather than the alternative Markdown, which enables us to specify custom CSS and JavaScript to integrate with the ScraperWiki platform.

ScraperWiki tools live in their own UNIX accounts called “boxes”, the code for the tool lives in a subdirectory, ~/tool, and web content in the ~/http directory is displayed. In this project the http directory contains a short JavaScript file, code.js, which by the magic of jQuery and some messy bash shell commands, puts the URL of the SQL endpoint into a file in the box. It also runs a package installation script once after the tool is first installed, the only package not already installed is the ggplot2 package.


function save_api_stub(){
scraperwiki.exec('echo "' + scraperwiki.readSettings().target.url + '" > ~/tool/dataset_url.txt; ')
}
function run_once_install_packages(){
scraperwiki.exec('run-one tool/runonce.R &> tool/log.txt &')
}
$(function(){
save_api_stub();
run_once_install_packages();
});

view raw

code.js

hosted with ❤ by GitHub

The ScraperWiki platform has an update hook, simply an executable file called update in the ~/tool/hooks/ directory which is executed when the underlying dataset changes.

This brings us to the meat of the viewer: the knitrview.R file calls the knitr package to take the view.Rhtml file and convert it into an index.html file in the http directory. The view.Rhtml file contains calls to some functions in R which are used to create the dynamic content.


#!/usr/bin/Rscript
# Script to knit a file 2013-08-08
# Ian Hopkinson
library(knitr)
.libPaths('/home/tool/R/libraries')
render_html()
knit("/home/tool/view.Rhtml",output="/home/tool/http/index.html")

view raw

knitrview.R

hosted with ❤ by GitHub

Code for interacting with the ScraperWiki platform is in the scraperwiki_utils.R file, this contains:

  • a function to read the SQL endpoint URL which is dumped into the box by some JavaScript used in the Rhtml template.
  • a function to read the JSON output from the SQL endpoint – this is a little convoluted since R cannot natively use https, and solutions to read https are different on Windows and Linux platforms.
  • a function to convert imported JSON dataframes to a clean dataframe. The data structure returned by the rjson package is comprised of lists of lists and requires reprocessing to the preferred vector based dataframe format.

Functions for generating the view elements are in view-source.R, this means that the R code embedded in the Rhtml template are simple function calls. The main plot is generated using the ggplot2 library. 


#!/usr/bin/Rscript
# Script to create r-view 2013-08-14
# Ian Hopkinson
source('scraperwiki_utils.R')
NumberOfTweets<-function(){
query = 'select count(*) from tweets'
number = ScraperWikiSQL(query)
return(number)
}
TweetsHistogram<-function(){
library("ggplot2")
library("scales")
#threshold = 20
bin = 60 # Size of the time bins in seconds
query = 'select created_at from tweets order by created_at limit 40000'
dates_raw = ScraperWikiSQL(query)
posix = strptime(dates_raw$created_at, "%Y-%m-%d %H:%M:%S+00:00")
num = as.POSIXct(posix)
Dates = data.frame(num)
p = qplot(num, data = Dates, binwidth = bin)
# This gets us out the histogram count values
counts = ggplot_build(p)$data[[1]]$count
timeticks = ggplot_build(p)$data[[1]]$x
# Calculate limits, method 1 – simple min and max of range
start = min(num)
finish = max(num)
minor = waiver() # Default breaks
major = waiver()
p = p+scale_x_datetime(limits = c(start, finish ),
breaks = major, minor_breaks = minor)
p = p + theme_bw() + xlab(NULL) + theme(axis.text.x = element_text(angle=45,
hjust = 1,
vjust = 1))
p = p + xlab('Date') + ylab('Tweets per minute') + ggtitle('Tweets per minute (Limited to 40000 tweets in total)')
return(p)
}

view raw

view-source.R

hosted with ❤ by GitHub

So there you go – not the world’s most exciting tool but it shows the way to make live reports on the ScraperWiki platform using R. Extensions to this would be to allow some user interaction, for example by allowing them to adjust the axis limits. This could be done either using JavaScript and vanilla R or using Shiny.

What would you do with R in ScraperWiki? Let me know in the comments below or by email: [email protected]

Jul 29 2013

pdftables – a Python library for getting tables out of PDF files

This post was first published at ScraperWiki.

One of the top searches bringing people to the ScraperWiki blog is “how do I scrape PDFs?” The answer typically being “with difficulty”, but things are getting better all the time.

PDF is a page description format, it has no knowledge of the logical structure of a document such as where titles are, or paragraphs, or whether it’s two column format or one column. It just knows where characters are on the page. The plot below shows how characters are laid out for a large table in a PDF file.

AlmondBoard7_LTChar

This makes extracting structured data from PDF a little challenging.

Don’t get me wrong, PDF is a useful format in the right place, if someone sends me a CV – I expect to get it in PDF because it’s a read only format. Send it in Microsoft Word format and the implication is that I can edit it – which makes no sense.

I’ve been parsing PDF files for a few years now, to start with using simple online PDF to text converters, then with pdftohtml which gave me better location data for text and now using the Python pdfminer library which extracts non-text elements and as well as bonding words into sentences and coherent blocks. This classification is shown in the plot below, the blue boxes show where pdfminer has joined characters together to make text boxes (which may be words or sentences). The red boxes show lines and rectangles (i.e. non-text elements).

AlmondBoard7

More widely at ScraperWiki we’ve been processing PDF since our inception with the tools I’ve described above and also the commercial, Abbyy software.

As well as processing text documents such as parliamentary proceedings, we’re also interested in tables of numbers. This is where the pdftables library comes in, we’re working towards making scrapers which are indifferent to the format in which a table is stored, receiving them via the OKFN messytables library which takes adapters to different file types. We’ve already added support to messytables for HTML, now its time for PDF support using our new, version-much-less-than-one pdftables library.

Amongst the alternatives to our own efforts are Mozilla’s Tabula, written in Ruby and requiring the user to draw around the target table, and Abbyy’s software which is commercial rather than open source.

pdftables can take a file handle and tell you which pages have tables on them, it can extract the contents of a specified page as a single table and by extension it can return all of the tables of a document (at the rate of one per page). It’s possible, for simple tables to do this with no parameters but for more difficult layouts it currently takes hints in the form of words found on the top and bottom rows of the table you are looking for. The tables are returned as a list of list of lists of strings, along with a diagnostic object which you can use to make plots. If you’re using the messytables library you just get back a tableset object.

It turns out the defining characteristic of a data scientist is that I plot things at the drop of a hat, I want to see the data I’m handling. And so it is with the development of the pdftables algorithms. The method used is inspired by image analysis algorithms, similar to the Hough transforms used in Tabula. A Hough transform will find arbitrarily oriented lines in an image but our problem is a little simpler – we’re interested in vertical and horizontal rows.

To find these vertical rows and columns we project the bounding boxes of the text on a page onto the horizontal axis ( to find the columns) and the vertical axis to find the rows. By projection we mean counting up the number of text elements along a given horizontal or vertical line. The row and column boundaries are marked by low values, gullies, in the plot of the projection. The rows and columns of the table form high mountains, you can see this clearly in the plot below. Here we are looking at the PDF page at the level of individual characters, the plots at the top and left show the projections. The black dots show where pdftables has placed the row and column boundaries.

AlmondBoard8_projection

pdftables is currently useful for supervised use but not so good if you want to just throw PDF files at it. You can find pdftables on Github and you can see the functionality we are still working on in the issue tracker. Top priorities are finding more than one table on a page and identifying multi-column text layouts to help with this process.

You’re invited to have a play and tell us what you think – [email protected]