<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Dragonfly Data Science - Medium]]></title>
        <description><![CDATA[Data science and dragonfly life - Medium]]></description>
        <link>https://medium.com/dragonfly-data-science?source=rss----e0ab7763f363---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Dragonfly Data Science - Medium</title>
            <link>https://medium.com/dragonfly-data-science?source=rss----e0ab7763f363---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 03:37:40 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/dragonfly-data-science" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Open science sprint]]></title>
            <link>https://medium.com/dragonfly-data-science/open-science-sprint-4e37edb6e4bd?source=rss----e0ab7763f363---4</link>
            <guid isPermaLink="false">https://medium.com/p/4e37edb6e4bd</guid>
            <dc:creator><![CDATA[Edward Abraham]]></dc:creator>
            <pubDate>Mon, 14 Jul 2014 05:09:48 GMT</pubDate>
            <atom:updated>2014-07-15T21:11:26.320Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eDPKy5iCFe5Db4kscZGZtQ.jpeg" /></figure><h4>Participate in the Mozilla open science software sprint, July 22nd and 23rd, 2014</h4><p>The Mozilla Science Lab is an organisation with the goal of building educational resources, tools and prototypes for the research community to make science more open, collaborative and efficient. Through the Software Carpentry project, they are coordinating a global programme to teach scientists to use software tools that promote shareable, reproducible research.</p><p>On July 22nd and 23rd, 2014, Science Lab are coordinating a global <a href="https://etherpad.mozilla.org/sciencelab-2014summersprint-sites">Open Science sprint</a>, to encourage people to take a couple of days out to promote reproducible research. Proposed projects include writing materials for introducing scientists to code review, software for extracting data from scientific literature, and development of training materials on data analysis.</p><p>At Dragonfly Data Science, some of us are planning to use the sprint as a spur to spend time on writing documentation and examples of how to do reproducible research, on packaging up software for release, and generally on contributing a little back to the open source projects that we use every day. One goal of ours is to contribute some examples to the scholarly markdown project (<a href="https://github.com/scholmd/scholmd">https://github.com/scholmd/scholmd</a>).</p><p>Alison Stringer is working with us to coordinate a Wellington event. We will host a distributed sprint, with a kickoff meeting on the morning of the 22nd and a debrief at the end of the 23rd. In between, we will work on projects in our own space and own time.</p><p>The focus of the sprint is as much on writing documentation and tutorials as it is on software development. It is suitable for anyone interested in promoting open science, even if they aren’t a software developer. If you would like to work on one of the sprint projects, would like to write a blog-post promoting open science, have some software that you would like to release, or would like to join in in some way, please get in touch with Alison Stringer (<a href="/service/mailto:alisonmstringer@gmail.com">alisonmstringer@gmail.com</a>).</p><p><a href="https://www.eventbrite.co.nz/e/welly-mozilla-open-science-software-sprint-tickets-12270961777">Sign up for the event here</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4e37edb6e4bd" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dragonfly-data-science/open-science-sprint-4e37edb6e4bd">Open science sprint</a> was originally published in <a href="https://medium.com/dragonfly-data-science">Dragonfly Data Science</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mirror your GitHub world to somewhere local]]></title>
            <link>https://medium.com/dragonfly-data-science/mirror-your-github-world-to-somewhere-local-23002a10aefc?source=rss----e0ab7763f363---4</link>
            <guid isPermaLink="false">https://medium.com/p/23002a10aefc</guid>
            <dc:creator><![CDATA[Christopher Knox]]></dc:creator>
            <pubDate>Thu, 03 Jul 2014 04:06:30 GMT</pubDate>
            <atom:updated>2014-07-03T04:33:22.413Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Q2PelgBxsDB0llblRPU-0w.jpeg" /></figure><h4>Backup all your GitHub code, so you can rest more easily</h4><p>We use <a href="http://www.github.com/">GitHub</a> for nearly all of our projects. All day people here are typing, and that content goes to live in the GitHub cloud. GitHub is a great service, but we feel more comfortable also having all our work backed up somewhere local.</p><p>For a while, we have run a local <a href="http://gitolite.com/gitolite/index.html">Gitolite </a>server, that we have tended to use for projects that don’t need all the GitHub features (and for clients who don’t want their projects living in the cloud). Recently, we started mirroring all our GitHub repositories to our own Gitolite server. We are now protected against the rogue account owner who inadvertently deletes all our things, against the sudden disappearance of github.com, or other unhappy accidents. As a bonus, we can now straightforwardly weed out inactive projects from our GitHub account when we like, knowing that they are archived. This means that we can start new projects without needing to upgrade our plan.</p><p>We automated this using an open-source BSD licensed python script (<a href="https://github.com/dragonfly-science/github-mirroring">available on GitHub of course</a>). The script has the following main options:</p><ul><li>mirror repositories from a user or an organization</li><li>mirror repositories to a Gitolite server (or to a local filesystem)</li><li>create a webhook on GitHub repositories to enable the script to be triggered automatically whenever repositories are updated</li></ul><p>The script could be run on occasion from the command line, or set up to run regularly. This might suit someone who wants to make sure all their code is copied across to their own computer. We make use of webhooks, so our continuous integration server is notified of any changes to the repositories. This ensures our mirror is always up to date. The process is strictly from GitHub to Gitolite and not in the other direction.</p><p>Please feel welcome to <a href="https://github.com/dragonfly-science/github-mirroring">check the project out </a>and make use of it. Issues and pull requests welcome.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=23002a10aefc" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dragonfly-data-science/mirror-your-github-world-to-somewhere-local-23002a10aefc">Mirror your GitHub world to somewhere local</a> was originally published in <a href="https://medium.com/dragonfly-data-science">Dragonfly Data Science</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Facebook’s emotional contagion: making something out of not much]]></title>
            <link>https://medium.com/dragonfly-data-science/facebooks-emotional-contagion-making-something-out-of-not-much-96364957330c?source=rss----e0ab7763f363---4</link>
            <guid isPermaLink="false">https://medium.com/p/96364957330c</guid>
            <dc:creator><![CDATA[Edward Abraham]]></dc:creator>
            <pubDate>Wed, 02 Jul 2014 06:19:21 GMT</pubDate>
            <atom:updated>2014-07-02T21:43:34.122Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/desat/multiply/grey/60/overlay/grey/1*JYuYugSiSZntt096elt4EA.jpeg" /></figure><h4>Your Facebook emotions may be contagious, but you’d be hard pressed to notice</h4><blockquote>We show, via a massive (<em>N </em>= 689,003) experiment on Facebook, that emotional states can be transferred to others via emotional contagion —<a href="http://www.pnas.org/content/111/24/8788.full">Kramer, Guillory, &amp; Hancock 2014</a></blockquote><p>Not only did the Facebook scientists manipulate the news feeds of their users, but their result were inconsequential. Sure, they dusted their paper with the trappings of significance testing: a smokescreen of <a href="http://www.nature.com/news/scientific-method-statistical-errors-1.14700"><em>P-</em>values</a> and so on. This gave the enterprise a suitable sciencey veneer, but when you have a sample size of 689,003 you run the risk that any old thing will be significant. This is the ‘<a href="http://galitshmueli.com/system/files/Print%20Version.pdf">too big to fail</a>’ problem with bells on.</p><p>The authors were coy about the effect size: a change in the proportion of positive or negative words of around one in a thousand. Not really enough to write home about, let alone to the Proceedings of the National Academy of Sciences.</p><p>Their treatment was to reduce the number of positive or negative posts that appeared in people’s news feeds. Their measure of the emotional response to this treatment was a count of the number of negative or positive words in subsequent posts (hopefully, no-one on social media uses sarcasm, right?). People who had positive posts removed from their feed reduced the proportion of positive words in subsequent posts from around 5.25% to 5.15%: a reduction in the proportion of positive words of around one word in a thousand. Rather than emotional contagion, a small effect like this may be something mundane, such as people talking about things other people have posted.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8jHFcEKjivzP5B5_efyuSg.jpeg" /><figcaption>The figure from the emotional contagion paper. Clipping the y-axis to make an effect appear larger is an old-fashioned sleight-of-hand that is poor science.</figcaption></figure><p>The main result is illustrated in their figure, which commits the graphical sin of truncating the bars. On the top side of the graph (positive words) the<em> y</em>-axis goes from 5.0% to 5.4%. If the bar started at zero, it would be ten times as high, and the responses would be so small they would be hard to see.</p><p>We don’t learn much from this paper, beyond that the effects are in the expected direction. Rather then solely going down the hypothesis testing route, science papers should tell us why their result matters. Crescendo statements such as ‘the well-documented connection between emotions and physical well-being suggests the importance of these findings for public health’ are not supported by this study. Given their remarkable data, and the disruption to their reputation caused by this study, it is a pity that they didn’t learn more from it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=96364957330c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dragonfly-data-science/facebooks-emotional-contagion-making-something-out-of-not-much-96364957330c">Facebook’s emotional contagion: making something out of not much</a> was originally published in <a href="https://medium.com/dragonfly-data-science">Dragonfly Data Science</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Don’t (re)draw with D3]]></title>
            <link>https://medium.com/dragonfly-data-science/dont-re-draw-with-d3-e396aa7de252?source=rss----e0ab7763f363---4</link>
            <guid isPermaLink="false">https://medium.com/p/e396aa7de252</guid>
            <dc:creator><![CDATA[Christopher Knox]]></dc:creator>
            <pubDate>Tue, 17 Jun 2014 01:56:20 GMT</pubDate>
            <atom:updated>2014-06-17T01:39:02.580Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/desat/multiply/blue/60/overlay/blue/1*ake_j6JMk8wch5gZ7ElkUg.png" /></figure><h4>Producing web-targeted graphics offline with minimal duplication of effort</h4><p>Producing a visualisation in R or Python for on-line interactive viewing commonly requires duplication of all rendering code and doesn&#39;t leverage web standards that simplify the problem. Here I will outline an alternative approach — which we used in the creation of <a href="http://www.mbie.govt.nz/what-we-do/business-growth-agenda/regions/interactive-web-tool">New Zealand’s 2014 Regional Economic Activity Report</a>.</p><p>The current approach to exporting graphics visualisation from an offline analysis and visualisation environment (<a href="http://ggplot2.org/">ggplot2</a>, <a href="http://matplotlib.org/">matplotlib</a>) generally consists of exporting a JSON representation of the image and its associated data to be loaded by a browser and drawn using <a href="http://d3js.org/">d3.js</a>. This requires both the implicit creation of a JSON schema for the representation of interactive graphics and the re-implementation all the rendering code. ‘All’ the rendering codes means not just the plotting, but also the axis, title, legend, and label layout — arguably a much larger component of good quality graphic production than the actual graphics. My experience is that this approach transforms an exciting interactive visualisation project into slavishly struggling to re-create the exact look of images as they are produced by an analyst. Fundamentally, the issue is that the code for creating good visualisations already exists; any web-based re-implementation is always going to be playing catchup.</p><p>The solution is to stop using D3 to draw. The first line of the D3 website&#39;s description of D3 states, “<strong>D3.js </strong>is a JavaScript library for manipulating documents based on data”. While D3 is often regarded as a visualisation or rendering library, on the homepage there is very little about graphics — the section headings are: ‘Selections’, ‘Dynamic properties’, ‘Enter and exit’, ‘Transformation not representation’, and ‘Transitions’. I believe these five headings provide a solid starting point for an interactive visualisations ‘grammar’ — unsurprisingly given the excellence of D3&#39;s design.</p><p>The plots below are time series created in ggplot2 with a custom theme, they are two of 2565 facets of dataset of employment across industries and regions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/430/1*n-qWirqkaW5ZO5q5hxUKDw.png" /><figcaption>Percentage employment in construction</figcaption></figure><p>The first plot shows the percentage employment in the construction sector at three different (related) regional levels.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/430/1*qVLT_UunV5-HSR4LzFmg8g.png" /><figcaption>Percentage employment in education and training</figcaption></figure><p>The second plot shows the same information for the education and training sector. The graphical differences between these plots are typical of the transitions required for an interactive graphic, but all that actually changes as you change sector are the values on the y-axis and the geometry of the lines themselves. Therefore, to support the ‘change-of-sector’ interaction in the browser a template image is required along with the y-axis labels and the line geometry for each sector.</p><p>Our workflow for the production of on-line interactive graphics has become:</p><ul><li>Save an image representing one facet of the image as an SVG file</li><li>Export the data needed to render the changes between facets</li><li>The browser renders the SVG file and we use D3 to manage the data binding and rendering associated with the facet-to-facet transitions and bind dynamic tool-tips and highlighting to the visualisation.</li></ul><p>Subsequent posts will address the technical details. In particular, I will demonstrate how to work around the current limitations surrounding SVG output, demonstrate how straightforward the required D3 code is, and show how to wire everything up into an <a href="http://emberjs.com/">ember.js</a> application.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e396aa7de252" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dragonfly-data-science/dont-re-draw-with-d3-e396aa7de252">Don’t (re)draw with D3</a> was originally published in <a href="https://medium.com/dragonfly-data-science">Dragonfly Data Science</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Make an Editor Like You]]></title>
            <link>https://medium.com/dragonfly-data-science/things-to-think-about-when-writing-reports-5fda1c030262?source=rss----e0ab7763f363---4</link>
            <guid isPermaLink="false">https://medium.com/p/5fda1c030262</guid>
            <dc:creator><![CDATA[Edward Abraham]]></dc:creator>
            <pubDate>Mon, 16 Jun 2014 21:44:00 GMT</pubDate>
            <atom:updated>2014-08-15T19:31:27.256Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*b0aFwM0FLTT1krtQQrQttw.jpeg" /></figure><h4>Things to think about when writing technical reports</h4><blockquote>A foolish consistency is the hobgoblin of little minds — Ralph Waldo Emerson</blockquote><p>When writing reports, let the persnickety hobgoblin thrive. Pay attention to the details. If you have addressed the details, the editor will appreciate your effort. And that can only help. Here are some common things to watch out for:</p><ul><li>Follow a consistent structure. In technical reports use standard sections such as Executive Summary, Introduction, Methods, Results, Discussion, Acknowledgements, and References. You may think there are reasons to stray from this boring sciencey structure, but usually people will find your work harder to follow as a consequence.</li><li>The topic sentence is your friend. The first sentence of every paragraph should be a sentence that concisely describes what the paragraph is about. Not many people will read your beautiful report from end to end. Well written topic sentences make life easier for skim readers, and will help you clarify your thinking.</li><li>Use short, active sentences. If your sentences have more than three commas, or are spread over more than three lines, they are likely to be too long. When you are revising your writing, look for sentences with plenty of clauses, and simplify them.</li><li>Your work was done in the past. To get the right tense, imagine that you are reading the report at some stage in the distant future. The present tense is used to refer to generally accepted truths, or to refer to other parts of the report. Otherwise, it’s past tense all the way.</li><li>Make sure that figures and tables are self-contained. The different colours and parts of the figure should be explained, all abbreviations should be spelt out, and quantities should have their units stated. You are trying to help the reader who gets the gist by hopping from figure to figure, without bothering with the text.</li><li>Use a spell checker. Yup, spell check before you share. No-one loves reading a report peppered with “teh”.</li><li>Never start a sentence with a number. Find a tidy way to rephrase it so the sentence starts with a word.</li><li>Watch your en-dashes. Dashes are used to indicate ranges. For example “1–10”, or “2001–02”. Don’t imagine for a moment no-one will notice that you are sneakily using a hyphen instead of a dash.</li><li>Expand acronyms, even common ones, on first use, i.e., “Heads-up Display (HUD)”. Since scuba is now a dictionary word, “Self-Contained Underwater Breathing Apparatus (SCUBA)” may be taking things one step too far.</li><li>Spell out numbers of things if the number is twelve or less, “there were two rabbits eating salad”, rather than “there were 3 horses up a tree”. The exception is when the number is part of a comparison with numbers over twelve, so “between 9 and 18 people in the lift”, rather than the clunky “between nine and 18&quot;.</li><li>Use standard abbreviations for units, so “the cheese weighed 20 kg”. Leave a space between the number and the abbreviated unit, and for extra points use a non-breaking space, so you avoid having an orphaned unit at the start of a line.</li></ul><p>If you are struggling with writing readable text, print out your document and take it somewhere away from the computer. Change your mode from keyboard mashing to reflection. Ideally, read the document out to yourself quietly, pen in hand. It can be surprising how tired eyes can jump over problems in the text that stand out clearly when you read it out loud.</p><p>If all else fails, follow the rules in the <a href="http://www.chicagomanualofstyle.org/">Chicago Manual Of Style</a>, 16th edition. This fabulous book is only over 1000-pages long if you include the indexes, and makes for a fun-filled read while you are planes, at the beach, or pretty much any time you have nothing better to be doing. You can learn all about the “Nine techniques for achieving gender neutrality” (p. 302), and become more hobgoblinish than the next person.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5fda1c030262" width="1" height="1" alt=""><hr><p><a href="https://medium.com/dragonfly-data-science/things-to-think-about-when-writing-reports-5fda1c030262">How to Make an Editor Like You</a> was originally published in <a href="https://medium.com/dragonfly-data-science">Dragonfly Data Science</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>