<?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[Stories by Steve Cotter on Medium]]></title>
        <description><![CDATA[Stories by Steve Cotter on Medium]]></description>
        <link>https://medium.com/@cottermedian?source=rss-de506226cbec------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*6Nu5BliP4PQdbpe4GRWHEw.jpeg</url>
            <title>Stories by Steve Cotter on Medium</title>
            <link>https://medium.com/@cottermedian?source=rss-de506226cbec------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 11:32:17 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@cottermedian/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Stop Mapping Counties, Start Mapping Exposure]]></title>
            <link>https://medium.datadriveninvestor.com/stop-mapping-counties-start-mapping-exposure-9aadd3ea0ead?source=rss-de506226cbec------2</link>
            <guid isPermaLink="false">https://medium.com/p/9aadd3ea0ead</guid>
            <dc:creator><![CDATA[Steve Cotter]]></dc:creator>
            <pubDate>Tue, 24 Mar 2020 03:01:47 GMT</pubDate>
            <atom:updated>2020-03-25T14:14:01.829Z</atom:updated>
            <content:encoded><![CDATA[<h4>At what point does public health supercede individual privacy?</h4><p><em>TL;DR — we should have a public dashboard that maps cases at the census block or block group level, and instances of potential public exposure identified through contact tracing at the address level. Something like South Korea, less the massive cell-phone surveillance and privacy intrusion. Identifying the locations of cases and places of potential exposure </em><strong><em>within a county</em></strong><em> would better communicate the proximity of the virus and may elicit more cooperation with stay-at-home and social distancing orders.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Us75psMc67gma5i7ERV2Ug.png" /><figcaption>Coronavirus (COVID-19) Cases in Pennsylvania as of March 23, 2020 from the <a href="https://www.health.pa.gov/topics/disease/coronavirus/Pages/Coronavirus.aspx">PA Department of Health</a></figcaption></figure><p>South Korea and China have effectively “<a href="https://www.nytimes.com/2020/03/23/world/asia/coronavirus-south-korea-flatten-curve.html">flattened the curve</a>” — at least thus far. Through aggressive early action, massive testing, contact tracing, and information dissemination, they have managed to beat back the virus.</p><p>One hallmark of South Korea’s approach is the use of technology to disseminate information on cases to all who may have been affected. <a href="https://www.nytimes.com/2020/03/23/world/asia/coronavirus-south-korea-flatten-curve.html">According to the New York Times</a>, mass messages are sent to everyone in the district when new cases are discovered. Apps and websites provide detailed timelines of infected people’s prior movements so citizens can cross-reference their own travel and get tested.</p><p><a href="https://www.datadriveninvestor.com/2018/03/22/big-money-and-cutting-edge-technology-how-investment-in-ai-ml-will-revolutionize-the-healthcare-industry/">Big Money and Cutting Edge Technology: How Investment in AI/ML Will Revolutionize The Healthcare Industry | Data Driven Investor</a></p><p>Contrast this with the United States. The first death in Allegheny County, Pennsylvania, where I live, was reported like this:</p><blockquote>The first COVID-19 death has occurred in Allegheny County, the Health Department has confirmed.</blockquote><blockquote>The only detail available is that the person was an adult who had been hospitalized.</blockquote><blockquote>“Saturday afternoon, Dan Laurent, spokesman for the Allegheny Health Network, confirmed that the patient who passed away from COVID-19 had been hospitalized at Jefferson Hospital but said <strong>no further information would be provided </strong>[emphasis added].”<br>- <a href="https://www.post-gazette.com/news/health/2020/03/21/First-Allegheny-County-COVID-19-death-reported/stories/202003210024">Pittsburgh Post Gazette - March 21, 2020</a></blockquote><p>Let me repeat that.</p><p>“<strong>No further information would be provided.”</strong></p><p>Allegheny County is 745 square miles. Where within the county did this patient live? Did they visit any public places in the days before their symptoms began or before they were hospitalized? Were they in front of you at the grocery last week?</p><p>Why are there no specifics? Wouldn’t concrete information help you to tie this to your everyday experience? Wouldn’t it help you to contextualize this information and really internalize that this <strong><em>is</em></strong> happening <strong><em>here</em></strong>? Knowing the total case count is great — but wouldn’t you rather know if you were potentially exposed? Why does a dead person’s privacy trump public safety?</p><p>The best data available on cases in the United States is from the <a href="https://www.nytimes.com/interactive/2020/world/coronavirus-maps.html">New York Times</a> and <a href="https://coronavirus.jhu.edu/map.html">John Hopkins University</a>. I have not found more detailed data than the county level (map below). While informative at the national level, county-level data is useless <strong>within a county</strong>. With little exception, such as someone infected at a specific university or company, this is the most detailed information being provided.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W9t5awv90x6CKcMrKtr-kQ.png" /><figcaption>County-Level Case Data from <a href="https://www.nytimes.com/interactive/2020/world/coronavirus-maps.html#us">The New York Times</a> as of March 23, 2020 at about 9:00 PM</figcaption></figure><p>The public is being asked — and in many cases being ordered— to stay-at-home, maintain social distance, and eliminate all-but-essential travel and social contact. Yet they are being provided highly abstract information based on sparse, even rationed, testing data. No wonder <a href="https://www.latimes.com/california/story/2020-03-22/california-beaches-coronavirus-stay-home-rules">they’re not listening</a>.</p><p>I postulate that providing more specific information to the public, as in South Korea, would increase compliance with stay-at-home orders and social-distancing by contextualizing the threat of the virus. Knowing that a confirmed case is in your county is very different from knowing that a confirmed case was at your grocery store. This could be achieved while respecting privacy rights.</p><p>Here are my suggestions:</p><p><strong>Public health departments and healthcare providers provide de-identified case data so it can be aggregated to the census block or block-group level.</strong> We‘re in the middle of the 2020 U.S. Census, which makes massive amounts of sensitive personal data available to the public by anonymizing it though aggregation. The same could be achieved with the domiciles of suspected or confirmed Covid-19 cases. Showing the locations of cases at a more local level may cause people to think twice before making unnecessary trips to the local grocery store or congregating with neighbors.</p><p><strong>Exposure in public locations identified through contact tracing should be entered into a centralized system for aggregation and public dissemination. </strong>Rather than leaving the public in the dark about their potential exposure, public health departments should make data on potential public exposure events available in aggregate for public places. For example, all patients being screened or tested for Covid-19 could be asked to complete a form listing the public places they visited — grocery stores, gas stations, parks, etc. — in the days before they became symptomatic. Providing this concrete information in aggregate across cases could help inform the public of potential exposure without infringing on the privacy of <a href="https://www.wired.com/story/phones-track-spread-covid19-good-idea/">all mobile phone users’ location data</a>.</p><p><strong>This is a simple app.</strong> Nothing described here is very complicated. A simple web application could facilitate the data collection and input by public health authorities and healthcare providers. A public dashboard could then be used to dissemination to the public similar to Johns Hopkins University’s Covid Tracker, but at a more detailed level. The Google Places API could be used to easily identify public places during contact tracing. The public could also be allowed to subscribe to notifications for their local area.</p><figure><img alt="A simple mockup of a form for collecting case information and potential public contact for public dissemination." src="https://cdn-images-1.medium.com/max/1024/1*oJbuKXXBlmU4KkLnSovdJA.png" /><figcaption>A simple mockup for a capturing case public contact history.</figcaption></figure><p><strong>But what about privacy and security? </strong>According to the HIPAA Rule, <a href="https://www.hhs.gov/hipaa/for-professionals/privacy/laws-regulations/index.html">“there are no restrictions on the use or disclosure of de-identified health information.”</a> So if data are sufficiently de-identified, HIPAA does not apply. Even if it did, many modern cloud application platforms like the <a href="https://cloud.google.com/security/compliance/hipaa/">Google Cloud Platform are already HIPAA compliant</a>. There’s also no need to collect Personally Identifiable Information (PII) in such a system. Patients could be asked to consent to sharing this information and be excluded if they did not consent. I’m betting that most would do so to protect and inform their fellow citizens.</p><p><strong>But getting the data is hard.</strong> The real challenge is gaining the cooperation of public health officials to provide the data. They often disclose very little information, citing privacy concerns. But if the information is sufficiently de-identified or aggregated, how is this an issue? It’s interesting that information of a very high public interest like potential public exposure to the virus is withheld, citing an individual patient’s right to privacy, while much broader rights— like the liberty to go outside — are suspended for everyone.</p><p><strong>Wouldn’t publishing data at the address level harm businesses?</strong> This is a matter of the greater good. While a business location with known exposure could see an effect on sales, it is more important that public and their employees know to that they may have been exposed and to watch for symptoms to protect other citizens and businesses. In many places, this argument is moot as only essential businesses are operating.</p><p><strong>What about the lack of testing? </strong>Due to scarcity and access issues around testing in the United States, the number of confirmed cases may vastly underestimate the current number of infections. Such a system could be used to collect suspected case data for individuals who may not have been tested or will never receive a test, but who have been asked to self-isolate by a healthcare professional. This may provide better data than the current system of counting only confirmed cases based on a test that is more readily available to the <a href="https://www.npr.org/sections/coronavirus-live-updates/2020/03/23/820455412/coronavirus-task-force-doctor-did-not-meet-guidelines-for-test-but-got-one-anywa">wealthy and politically connected</a> than sick people.</p><p><strong>Couldn’t the public self-report their own cases to track the spread rather than a public health official or healthcare provider? </strong>The current testing system is very restrictive unless you’re <a href="https://www.npr.org/sections/coronavirus-live-updates/2020/03/23/820455412/coronavirus-task-force-doctor-did-not-meet-guidelines-for-test-but-got-one-anywa">wealthy or politically connected</a>. Thus any public reporting system based upon it may underestimate the number of infections and misstate the virus’ spread. So there is a legitimate argument that allowing the public to self-report cases, instead of a public health official or healthcare provider, could provide a better indication of the true spread. But a self-reporting system is more prone to manipulation through bots, trolls, etc. It would also be difficult to de-identify such data and ensure data integrity.</p><p><strong>Could more detailed information do more harm than good?</strong> Would an unintended consequence of more information cause the public to be less cautious? It’s hard to know, but I’d argue that the current system of county-level case counts already does just that.</p><p>Given <a href="https://spcotter.github.io/">my prior work</a> in data collection and visualization, particularly with maps, my colleagues and I at <a href="https://kinertia.com/">Kinertia</a> discussed building such a system. But we concluded that buy-in from a local government or public health authority to provide the data would first be necessary.</p><p>Any volunteers?</p><p>Are you frustrated by the lack of local data and information on the Covid 19 pandemic? Are you tired of the current information dissemination system of near-daily press conferences from every level of government? Can you think of more data-driven approaches to keeping the public informed and engaged in social distancing while also respecting individual privacy?</p><p>I am not an epidemiologist or a a public health expert. But I do have a background in public policy, data analysis, and software development. And personally, I’d like to know much more about how and where this virus is spreading to better inform my decisions.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fb2a0d6%3Fas_embed%3Dtrue&amp;dntp=1&amp;display_name=Upscribe&amp;url=https%3A%2F%2Fupscri.be%2Fb2a0d6%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href">https://medium.com/media/0707f5c806284d01a4a13c7b13a91ce3/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9aadd3ea0ead" width="1" height="1" alt=""><hr><p><a href="https://medium.datadriveninvestor.com/stop-mapping-counties-start-mapping-exposure-9aadd3ea0ead">Stop Mapping Counties, Start Mapping Exposure</a> was originally published in <a href="https://medium.datadriveninvestor.com/">DataDrivenInvestor</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Quarter-of-a-Million Healthy Rides]]></title>
            <link>https://medium.com/@cottermedian/a-quarter-of-a-million-healthy-rides-c30c571cc8ff?source=rss-de506226cbec------2</link>
            <guid isPermaLink="false">https://medium.com/p/c30c571cc8ff</guid>
            <dc:creator><![CDATA[Steve Cotter]]></dc:creator>
            <pubDate>Tue, 18 Jun 2019 02:54:16 GMT</pubDate>
            <atom:updated>2019-06-18T02:54:16.185Z</atom:updated>
            <content:encoded><![CDATA[<h4>A Visualization of Pittsburgh’s Bike Share System</h4><p>Pittsburgh’s Bike Share System, <a href="https://healthyridepgh.com/">Healthy Ride</a>, launched in June of 2015. By June 30, 2018, riders had made 238,467 trips. The network currently includes 52 stations and <a href="https://healthyridepgh.com/expansion/">continues to expand</a>. With all this data, I just had to <a href="https://spcotter.github.io/healthy-ride-pgh/">build something.</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NXlmOtq-F4PV0fpk0wpxPg.png" /><figcaption>An animation of ride arcs over time.</figcaption></figure><p>I first experimented with this data by <a href="https://spcotter.github.io/healthy-ride-pgh/">animating ride arcs between stations</a> in proportion to how long the ride actually took. But that was many years ago, and now there are about six times the number of trips in the available data. What can we learn about how people use the bike share program from all this information?</p><h3>Stations</h3><p>First, let’s map the 52 bike share stations which are mostly are located in the city’s eastern neighborhoods.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pRnt5bUyGCAUJcW9prw27w.png" /><figcaption>Pittsburgh’s Bike Share network comprises 52 stations.</figcaption></figure><p>Street maps fail to capture the topography of the city, which has a lot of hills. Many are quite steep. Adding topography to our map helps to explain the geographic distribution of the stations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zl4o_pdRntpPtdf92cBwQA.png" /><figcaption>Pittsburgh has a lot of hills.</figcaption></figure><h3>Trips</h3><p>Let’s now look at trips. Which stations are busiest? It’s easy enough to convert our station map in a bubble plot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*okYA_S4RUCkqu_HCM1cXQQ.png" /><figcaption>A bubble plot of stations sized by their number of trips.</figcaption></figure><p>Here we can see that the busiest stations are on the Southside, Downtown, and in the Strip District. In fact, the busiest station is <strong>Southside Works</strong>, with 12,641 trips or about 5.3 percent of all trips, followed closely by <strong>Liberty and Stanwix</strong>, with 12,021 trips, and <strong>Market Square</strong> with 11,709. As we can see, these areas are very flat, making them more bike-friendly. What we can’t see in this particular map is that they are also very close to bike lanes and trails.</p><p>The least busy station is near the center of the map on aptly named <strong>Centre Ave &amp; Kirkpatrick St</strong>. This station is isolated by its surrounding terrain in the middle of the Hill District. While it may not be as busy as the tourist-filled areas of Market Square, with 662 trips, it is certainly being used and may be a vital connection for residents who are up for the 250 foot ascent back up the hill.</p><h3>The Network</h3><p>The trip data include the starting stations and the return station for each trip, forming a directed graph. Let’s add the edges to our bubble plot above. Here, the opacity and weight of each line is scaled to the number of trips between those stations. To incorporate trips that start and end at the same station, I also shaded and weighted the border of each station circle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RjiPXcBZGnxBO9MXbK2wTQ.png" /><figcaption>A visualization of bike share trips between stations, forming a directed graph.</figcaption></figure><p>Now we can see which stations are related based on the trips between each pair. Downtown has strong connections to the North Side, the Strip District, Southside, and Lawrenceville. Oakland, Shadyside, and Bakery Square have strong interconnections, as do Shadyside, Bloomfield, and Lawrenceville.</p><p>Of all trips, 26.7 percent or 63,862 began and ended at the same station. Looking at this statistic for individual stations reveals an interesting distinction. Some stations, including the busiest, are what I’ll call <strong><em>out-and-back</em></strong> stations — the bike is more likely to be returned to the same station than to another. I’ll call the others <strong><em>commuter</em></strong> stations — the bike is more likely to be returned to a different station. The vast majority of stations — 42 of 52 — are <strong><em>out-and-back</em></strong> stations. Only 10 are <strong><em>commuter</em></strong> stations. That’s interesting. So which stations are these?</p><h4>Fifth Ave &amp; S Bouquet St <strong>Commuter</strong></h4><p><em>6,309 rides, 2.6% of total</em></p><p>The most popular destination station from here is <strong>Boulevard of the Allies &amp; Parkview Ave </strong>with 18.8 percent of trips taking 14 minutes on average. These are likely students riding from their apartments in South Oakland to class and back.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*73YE204rl5GucYyBmnWD_A.png" /><figcaption>Fifth Ave &amp; S Bouquet St</figcaption></figure><h4>17th St &amp; Penn Ave</h4><p><em>5,980 rides, 2.5% of total</em></p><p>The David L. Lawrence Convention Center is the most popular return station from here with 16.6 percent of trips. These bikers are out joy-riding with an average trip duration of 30 minutes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aaUXWYoqJYC4F1snsRscPw.png" /><figcaption>17th St &amp; Penn Ave</figcaption></figure><h4>Boulevard of the Allies &amp; Parkview Ave</h4><p><em>5,265 rides, 2.2% of total</em></p><p>Here go those students again. They must be late for class with 25 percent of trips heading back up to Fifth Ave &amp; S Bouquet St in only 9 minutes, compared to the 14 minutes on the way down.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7ZiFQjWLKbEY23FSut9aLQ.png" /><figcaption>Boulevard of the Allies &amp; Parkview Ave</figcaption></figure><h4>Ross St &amp; Sixth Ave (Steel Plaza T Station)</h4><p><em>4,354 rides, 1.8% of total</em></p><p>As the last of the 10 commuter stations I profile, the most popular destination is S 12th St &amp; E Carson St on the Southside taking an average of 25 minutes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1qB6jGdgW8c73-9GuWeWaQ.png" /><figcaption>Ross St &amp; Sixth Ave (Steel Plaza T Station)</figcaption></figure><p>To find clusters of stations quantitatively, I applied <a href="https://en.wikipedia.org/wiki/Spectral_clustering">Spectral Clustering</a> using the proportion of trips between each pair of stations as the measure of similarity. In the map above, imagine that the lines between stations are actually strings. Stations with more trips have thicker strings strung between them. Spectral clustering tells us how to divide up the stations into a given number of clusters so that we cut the fewest and thinnest strings. This reveals the clusters of stations formed by trips.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x0zbGW8NTvJbHhONAmltig.png" /><figcaption>Results of Spectral Clustering with six clusters.</figcaption></figure><p>With the number of clusters set to six, we have three large clusters — 17 stations combining Shadyside with East Liberty and Bloomfield; 15 stations comprising the North Side combined with northern Downtown and Lawrenceville; 13 stations from the Southside grouped with the southern stations Downtown, and the southeastern stations in Oakland. We also have one smaller cluster of 5 stations in Oakland and two stations that are each in their own cluster, the <strong>Healthy Ride Hub</strong> and <strong>Centre Ave &amp; Kirkpatrick St</strong>.</p><p>If you’d like to explore the data further, check out the <a href="https://spcotter.github.io/healthy-ride-pgh/">interactive visualization</a>. It’s written in React.js and uses the wonderful Leaflet.js library for mapping. I crunched the numbers in R. All the code is <a href="https://github.com/spcotter/healthy-ride-pgh">available on GitHub</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*13W_G4eY93FJiWVgvWZIfg.png" /><figcaption>An interactive visualization of Pittsburgh’s Bike Share System.</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c30c571cc8ff" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Hate Hits Home]]></title>
            <link>https://medium.com/@cottermedian/hate-hits-home-becfaf007f8f?source=rss-de506226cbec------2</link>
            <guid isPermaLink="false">https://medium.com/p/becfaf007f8f</guid>
            <category><![CDATA[mass-shootings]]></category>
            <category><![CDATA[hate-crimes]]></category>
            <category><![CDATA[pittsburgh]]></category>
            <dc:creator><![CDATA[Steve Cotter]]></dc:creator>
            <pubDate>Sat, 27 Oct 2018 22:34:40 GMT</pubDate>
            <atom:updated>2018-10-27T22:34:40.180Z</atom:updated>
            <content:encoded><![CDATA[<h4>Smaller cities like Pittsburgh often enjoy national and international media attention. But not like this.</h4><p>Today — October 27, 2018 — a man entered a synagogue in Pittsburgh and began killing people because they are Jewish. There are now at least eleven people dead.</p><p>This atrocity was the top story on many news websites both nationally and internationally. Some might say that the media attention paid to such mass shootings is <a href="http://www.latimes.com/nation/la-na-orlando-shooting-media-20160618-snap-story.html">irresponsible and contributes to the problem</a>. I tend to see the international coverage of such incidents as evidence that people across the world are sickened by such hate.</p><p>Below is a collection of screenshots of major media sites locally, nationally, and internationally from this afternoon. It is not my intention to make any sort of statement with them. They are only an observation of how the events of today were covered by about three-dozen media websites.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*vwKZWVAAAOHah57W8TfdHQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*HmA9vKSEqf9GHVDKdK6jNw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*LuxfLejablmYcC9KtzObsg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*UTJYEKrpWW_IOu2AjzWfEA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*vFLGginoVsu-4UUJshT5hA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*qfHv72Emc1hniqWeaMn29Q.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*-n3xZYyWN2z5PZrlN99mRw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*P4fgIOI5JgOQcqp9IQ3bnw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*AXmEfd1MVDTlmvO4zQFvcw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*mV7uQJxAKhsLjGIsZYslqA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*JVOzNlrIj3vFWoBQ2JLaUw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*dZzMnajVE2i5NTsc1BMpag.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*Rbdo_LMdBsZezQlU5qxNDw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*ftD_7v_AWKtwkizk91jmCw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*3_mmM8xERNzfh58QpX1t2A.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*47wN9vifdi9eYdlbFtnDOQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*Rd8p6BtcYfK7s10u1DDN1g.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*v0eiUx1rU4DLI5Sm-pTj6w.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*_JVz-xp3hSQMkfEM6QFu1w.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*o7Ad1MYSsumWTan53YyU0w.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*lB6y_z_WNzIKdiQqHgFhDQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*WDcqwc4ktF9BhubmUs1Z6A.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*k67tNVj3jaGqk_4p_d-pDQ.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*KwY28aoabWrk_GtS254Fog.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*bzEPofCp_y8Mm7kjr9A6Kw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*7V-92ThmadVMux2clhAOHA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*hlpYBICQwnKFjz2IMRSPtw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*4l_7mXv-AcII8YLPr9UOEw.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*fKfygxBWSemXuvaN75hI6w.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/1*JhKR1bBkSCgsnnpuQ8KZyg.jpeg" /></figure><h4>Sources</h4><p><a href="https://mobile.lemonde.fr/ameriques/article/2018/10/27/etats-unis-plusieurs-victimes-dans-une-fusillade-pres-d-une-synagogue-a-pittsburgh_5375554_3222.html">https://mobile.lemonde.fr/ameriques/article/2018/10/27/etats-unis-plusieurs-victimes-dans-une-fusillade-pres-d-une-synagogue-a-pittsburgh_5375554_3222.html</a></p><p><a href="https://www.washingtonpost.com/nation/2018/10/27/pittsburgh-police-responding-active-shooting-squirrel-hill-area/?utm_term=.a482e926638e">https://www.washingtonpost.com/nation/2018/10/27/pittsburgh-police-responding-active-shooting-squirrel-hill-area/?utm_term=.a482e926638e</a></p><p><a href="https://www.nytimes.com/2018/10/27/us/active-shooter-pittsburgh-synagogue-shooting.html">https://www.nytimes.com/2018/10/27/us/active-shooter-pittsburgh-synagogue-shooting.html</a></p><p><a href="https://www.wpxi.com/news/top-stories/suspect-identified-in-mass-shooting-at-tree-of-life-synagogue-in-pittsburgh/860969031">https://www.wpxi.com/news/top-stories/suspect-identified-in-mass-shooting-at-tree-of-life-synagogue-in-pittsburgh/860969031</a></p><p><a href="https://www.wtae.com/article/squirrel-hill-synagogue-shooting-tree-of-life-congregation-pittsburgh/24316467">https://www.wtae.com/article/squirrel-hill-synagogue-shooting-tree-of-life-congregation-pittsburgh/24316467</a></p><p><a href="https://pittsburgh.cbslocal.com/2018/10/27/heavy-police-presence-near-synagogue-in-squirrel-hill/">https://pittsburgh.cbslocal.com/2018/10/27/heavy-police-presence-near-synagogue-in-squirrel-hill/</a></p><p><a href="https://www.bbc.com/news/world-us-canada-46002549">https://www.bbc.com/news/world-us-canada-46002549</a></p><p><a href="https://www.usatoday.com/story/news/2018/10/27/pittsburgh-police-report-active-shooter-near-synagogue/1788556002/">https://www.usatoday.com/story/news/2018/10/27/pittsburgh-police-report-active-shooter-near-synagogue/1788556002/</a></p><p><a href="https://www.foxnews.com/us/pittsburgh-police-confirms-active-shooter-at-synagogue-multiple-victims-reported">https://www.foxnews.com/us/pittsburgh-police-confirms-active-shooter-at-synagogue-multiple-victims-reported</a></p><p><a href="https://www.nbcnews.com/news/us-news/active-shooter-reported-near-pittsburgh-synagogue-n925211">https://www.nbcnews.com/news/us-news/active-shooter-reported-near-pittsburgh-synagogue-n925211</a></p><p><a href="https://www-m.cnn.com/2018/10/27/us/pittsburgh-synagogue-active-shooter/index.html?r=https%3A%2F%2Fwww.cnn.com%2F">https://www-m.cnn.com/2018/10/27/us/pittsburgh-synagogue-active-shooter/index.html?r=https%3A%2F%2Fwww.cnn.com%2F</a></p><p><a href="https://m.huffpost.com/us/entry/us_5bd475ffe4b055bc948c6ef2">https://m.huffpost.com/us/entry/us_5bd475ffe4b055bc948c6ef2</a></p><p><a href="https://www.wsj.com/articles/pittsburgh-authorities-report-active-shooter-1540653056?mod=mhp">https://www.wsj.com/articles/pittsburgh-authorities-report-active-shooter-1540653056?mod=mhp</a></p><p><a href="http://www.latimes.com/nation/nationnow/la-na-pittsburgh-synagogue-20181027-story.html">http://www.latimes.com/nation/nationnow/la-na-pittsburgh-synagogue-20181027-story.html</a></p><p><a href="https://nypost.com/2018/10/27/shooter-screamed-all-jews-must-die-before-opening-fire-at-pittsburgh-synagogue/">https://nypost.com/2018/10/27/shooter-screamed-all-jews-must-die-before-opening-fire-at-pittsburgh-synagogue/</a></p><p><a href="https://www.theguardian.com/us-news/2018/oct/27/pittsburgh-synagogue-shooting">https://www.theguardian.com/us-news/2018/oct/27/pittsburgh-synagogue-shooting</a></p><p><a href="https://abcnews.go.com/US/police-responding-active-shooter-pittsburgh-synagogue/story?id=58790381&amp;cid=clicksource_4380645_1_hero_headlines_bsq_hed">https://abcnews.go.com/US/police-responding-active-shooter-pittsburgh-synagogue/story?id=58790381&amp;cid=clicksource_4380645_1_hero_headlines_bsq_hed</a></p><p><a href="https://www.cbsnews.com/news/pittsburgh-shooting-today-tree-of-life-synagogue-squirrel-hill-pennsylvania-2018-10-27-live-updates/">https://www.cbsnews.com/news/pittsburgh-shooting-today-tree-of-life-synagogue-squirrel-hill-pennsylvania-2018-10-27-live-updates/</a></p><p><a href="https://www.yahoo.com/news/live-updates-shooting-pittsburgh-synagogue-153125806.html">https://www.yahoo.com/news/live-updates-shooting-pittsburgh-synagogue-153125806.html</a></p><p><a href="https://pittsburgh.cbslocal.com/2018/10/27/heavy-police-presence-near-synagogue-in-squirrel-hill/">https://pittsburgh.cbslocal.com/2018/10/27/heavy-police-presence-near-synagogue-in-squirrel-hill/</a></p><p><a href="https://www.chicagotribune.com/news/nationworld/ct-pittsburgh-synagogue-active-shooter-20181027-story.html#nt=screamer">https://www.chicagotribune.com/news/nationworld/ct-pittsburgh-synagogue-active-shooter-20181027-story.html#nt=screamer</a></p><p><a href="https://m.sfgate.com/news/crime/article/Pittsburgh-police-report-active-shooter-near-13341522.php">https://m.sfgate.com/news/crime/article/Pittsburgh-police-report-active-shooter-near-13341522.php</a></p><p><a href="https://www.npr.org/2018/10/27/661347236/multiple-casualties-in-shooting-near-pittsburgh-synagogue">https://www.npr.org/2018/10/27/661347236/multiple-casualties-in-shooting-near-pittsburgh-synagogue</a></p><p><a href="http://time.com/5436829/multiple-casualties-shooting-pittsburgh-synagogue/">http://time.com/5436829/multiple-casualties-shooting-pittsburgh-synagogue/</a></p><p><a href="https://www.thetimes.co.uk/edition/news/eight-people-killed-in-pittsburgh-tree-of-life-synagogue-shooting-59sq69hmh">https://www.thetimes.co.uk/edition/news/eight-people-killed-in-pittsburgh-tree-of-life-synagogue-shooting-59sq69hmh</a></p><p><a href="https://www.dailymail.co.uk/news/article-6323781/Active-shooter-Pittsburgh-synagogue.html">https://www.dailymail.co.uk/news/article-6323781/Active-shooter-Pittsburgh-synagogue.html</a></p><p><a href="https://www.bostonglobe.com/news/nation/2018/10/27/pittsburgh-police-report-active-shooter-near-synagogue/Vc5Wno0znEoPotQ0JtN25J/story.html">https://www.bostonglobe.com/news/nation/2018/10/27/pittsburgh-police-report-active-shooter-near-synagogue/Vc5Wno0znEoPotQ0JtN25J/story.html</a></p><p><a href="https://apple.news/AULOrVO_2RYa_rA6CbwNVNA">https://apple.news/AULOrVO_2RYa_rA6CbwNVNA</a></p><p><a href="https://torontosun.com/news/crime/active-shooter-multiple-casualties-after-shooting-near-pittsburgh-synagogue/wcm/a12f0d7a-30fd-4342-adfa-ce97e11885c3">https://torontosun.com/news/crime/active-shooter-multiple-casualties-after-shooting-near-pittsburgh-synagogue/wcm/a12f0d7a-30fd-4342-adfa-ce97e11885c3</a></p><p><a href="https://www.cbc.ca/news/world/pittsburgh-active-shooter-near-synagogue-1.4881234">https://www.cbc.ca/news/world/pittsburgh-active-shooter-near-synagogue-1.4881234</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=becfaf007f8f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[“Machine Learning” should be called “Blind Optimization”]]></title>
            <link>https://medium.datadriveninvestor.com/machine-learning-should-be-called-blind-optimization-928f20f5fe6?source=rss-de506226cbec------2</link>
            <guid isPermaLink="false">https://medium.com/p/928f20f5fe6</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[anthropomorphism]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Steve Cotter]]></dc:creator>
            <pubDate>Tue, 23 Oct 2018 04:10:37 GMT</pubDate>
            <atom:updated>2018-11-05T14:37:29.854Z</atom:updated>
            <content:encoded><![CDATA[<h4>The way we anthropomorphize optimization algorithms as “intelligent” systems that “learn” and “think” mystifies the public while degrading real human intelligence. Never mind the arrogance.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*c6rS8tc-G4xjxiAS" /><figcaption>by <a href="https://unsplash.com/@markusspiske?utm_source=medium&amp;utm_medium=referral">Markus Spiske</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Computer scientists like to anthropomorphize their creations, often equating human thought to computation. How rampant is this reductionism? It’s in the very name of the fields of Artificial <em>Intelligence</em> and Machine <em>Learning</em>. These terms have a very technical meaning known to experts. But tools from these fields are increasingly being applied in areas where non-experts need to be well-informed and need to have a say in how and when they’re applied.</p><p>The language we use to describe things matters immensely when we’re trying to understand them. For example, when discussing evolution, we often hear that an organism “has evolved” some ability. This is a terrible way to describe evolution. It’s so terrible, it actually poisons understanding. Describing the process in this way — like it’s something that was done BY an organism — implies the organism did something to evolve. But an organism actively evolves no more than you actively gravitate toward the earth when you fall from a ladder. Evolution is something that happens TO an organism in the same way that gravity happens TO you. They have no agency or say in the process. Casually speaking about evolution like this gives non-experts the wrong idea about what’s really going on. They form an incorrect mental model of the process that can be difficult to correct.</p><p>Casually assigning human-like qualities like “intelligence” and “learning” to algorithms, and speaking about them like they “think” and have agency has the same effect when discussing Machine Learning. It gives the massively wrong idea to non-experts as to what’s really going on. It leaves them with the impression that we’ve created a human-level intelligence. It puts a picture of a human brain in their heads and encourages them to infer other human-like qualities that algorithms cannot possess — qualities like common sense or morality. This poisons their ability to make informed decisions about the responsibilities we, as a society, should assign to these algorithms.</p><p>Let’s pretend you know nothing of AI/ML. You’re outside on your front street teaching your child to ride a bike on the sidewalk. Someone stops by and asks if you’d mind if their newest self-driving car can practice on the street too. If they describe it to you as being driven by “Artificial Intelligence and Machine Learning” you might infer that it’s intelligent enough to not run over your child. But if they instead tell to you that it’s actually an algorithm that takes hundreds or thousands of inputs from dozens of sensors, cameras, and LIDAR and combines them in an very complex algorithm optimized to steer, accelerate, and brake based on previous readings and data, you’re much more likely to ask if their previous data include children on bicycles. (It’s important to note that the approach thus far has been that the data do not include children with bicycles and they’re driving down your street to collect it.)</p><p>Machine Learning algorithms are blind optimization. Calling them what they are cuts through the aspirational jargon to give non-experts a much better idea of what’s really going on. The algorithms are blind in the sense that they have no insight into the true goals of their creators. They are computer programs that crank through computations optimized for the data they were built on. They are no more, no less. They can be immensely complex with nodes and layers galore. But complexity is not intelligence. Discussing them like they’re something greater, something nearly human, seems to me to be irresponsible, maybe even unethical. Has the lofty language we’ve used for these algorithms influenced how we think about them and how skeptical we are of their capabilities?</p><p>Maybe “Blind Computation” is too harsh. But the point holds true. Describing the whole field from the perspective of the Machine, the “learner”, removes the most important element — the <strong>teacher</strong>. The teacher is ultimately responsible for the decisions of the machine, whether it erroneously drives up on the sidewalk, <a href="https://www.theguardian.com/technology/2018/mar/19/uber-self-driving-car-kills-woman-arizona-tempe">fails to recognize a pedestrian</a>, or <a href="https://www.ted.com/talks/james_bridle_the_nightmare_videos_of_childrens_youtube_and_what_s_wrong_with_the_internet_today">recommends a violent video to a child</a>. So instead of “Blind Computation” how about “<a href="https://www.linkedin.com/pulse/machine-teaching-vs-learning-philippe-lambinet">Machine Teaching</a>” to reinforce the role of humans, our role, and, most importantly, our responsibility in giving algorithms such power?</p><p>I am by no means an expert in Machine Learning. But with a background in Statistics and Computation, I’m informed enough to sigh hopelessly every time I hear that the latest adversarial DNN has spawned the singularity.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=928f20f5fe6" width="1" height="1" alt=""><hr><p><a href="https://medium.datadriveninvestor.com/machine-learning-should-be-called-blind-optimization-928f20f5fe6">“Machine Learning” should be called “Blind Optimization”</a> was originally published in <a href="https://medium.datadriveninvestor.com/">DataDrivenInvestor</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Visualizing Pittsburgh Transit]]></title>
            <link>https://medium.com/@cottermedian/visualizing-pittsburgh-transit-1ab05664b8b3?source=rss-de506226cbec------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ab05664b8b3</guid>
            <category><![CDATA[animation]]></category>
            <category><![CDATA[transportation]]></category>
            <category><![CDATA[pittsburgh]]></category>
            <category><![CDATA[public-transit]]></category>
            <category><![CDATA[data-visualization]]></category>
            <dc:creator><![CDATA[Steve Cotter]]></dc:creator>
            <pubDate>Wed, 14 Sep 2016 01:44:26 GMT</pubDate>
            <atom:updated>2016-09-14T02:05:32.535Z</atom:updated>
            <content:encoded><![CDATA[<h4>What does 6,500 trips look like?</h4><p>The buses are slow. They’re always late and overcrowded. They don’t run often enough. The experience of Port Authority can leave you with complaints like these. But when you’re riding, or more likely waiting for, the bus or T, do you ever consider how large and complex our transit system is?</p><p>Port Authority makes over 6,500 trips every weekday, racking up 213,000 rides. That’s like moving nearly 20% of the entire county’s population every day. How can you fathom 6,500 trips per day? What does that look like? Let’s find out.</p><p>Many transit systems publish their schedule data in the <a href="https://developers.google.com/transit/gtfs/">General Transit Feed Specification</a> (GTFS) format. These files describe all the system’s routes, stop locations, trips, and stop times. Mapping services like Google use these data to provide transit directions.</p><p>First, consider the routes. Port Authority has over 300. GTFS data include a shapes table of the latitude and longitude point sequence for each route. To plot these, I used PostGIS to convert them into lines and exported them to GeoJSON. Using Leaflet and D3.js, <a href="https://spcotter.github.io/visualizing-pittsburgh-transit/routes.html">I mapped them</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8Wcdpt_Mp5ZKWKy6-ltYGw.png" /><figcaption>Port Authority’s routes based on GTFS data. Busway, HOV, and light rail routes are color-coded. Routes are slightly transparent, so darker lines are overlapping routes.</figcaption></figure><p>Mapping the routes conveys the extent of the system and highlights the busway and light-rail routes. Routes are <a href="https://en.wikipedia.org/wiki/List_of_bus_routes_in_Pittsburgh#Route_renumbering_and_realignments">color-coded</a>. Those that begin with “P” and operate on the East Busway are shaded purple. “Y” routes are yellow for the South Busway. And so on.</p><p>The route map also illustrates Port Authority’s direct service model, which is <a href="http://usa.streetsblog.org/2011/06/20/profiles-in-american-brt-pittsburghs-south-busway-and-east-busway/">pretty unique</a>. Notice how far out many of the busway routes go. As they enter the urban core, suburban express routes jump on the busways to bypass congestion and go all the way downtown. There’s no need to transfer.</p><p>Now that we have the routes, we need the buses — trips in GTFS lingo. The data provide the stop coordinates and times for each trip. So at any given moment, if a bus has a scheduled stop you can easily tell where it is.</p><p>Now comes the tricky part. When a bus isn’t at a stop, it’s obviously somewhere between. But where? I made the assumption buses proceed at a constant rate between stops. So, if a bus stops at Point A at 9:00 AM and Point B at 9:10 AM, I assume that at 9:05 it’s halfway between A and B along it’s route. This is simple linear interpolation.</p><p>The SQL is not so simple. For buses that don’t have stops at the given moment, I find the previous and next stops. Then, I find the points along the route that are closest to the stop coordinates and cut the route line between these points. Finally, I interpolate a point along the route substring based on the fraction of minutes that have passed since the last stop. Rather than trying to do all of this at once, I wrote <a href="https://github.com/spcotter/visualizing-pittsburgh-transit/blob/master/interpolated_route_points_at_time.sql">the query</a> for a particular time of day. With a simple Ruby script, I ran it for each minute of each hour of the day. Finally, I used Node.js to transform the output into a smaller data format.</p><p>With the locations of all the active buses at each minute, I added them to the map.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fFR5ILQBjSXu9fImEfQ99g.png" /><figcaption>The scheduled location of Port Authority’s weekday trips at 9:00 AM based on GTFS data. Busway, HOV, and light-rail routes and trips are color-coded.</figcaption></figure><p>Now let’s animate the whole system and see what 6,500 trips look like. While D3.js seems like the obvious choice, it turns out that animating thousands of SVG elements on the page will crush your browser. The canvas element with requestAnimationFrame is a better choice.</p><p>Using the canvas introduces another complication — the animation is way too fast. Most browsers animate at around 60 frames per second. At that speed, animating the entire weekday would take 24 seconds. But if we slow things down too much, the animation won’t be smooth. To address this, I added more interpolation in JavaScript. But this comes with a tradeoff.</p><p>We only know the locations of the buses on the routes at one-minute intervals. Interpolating between those intervals means our buses won’t necessarily stay on the routes. At low zoom levels, this won’t be very noticeable. But zoomed-in, our buses will deviate from the route lines. The only way to fix this is to run the visualization at a higher resolution than one minute. But the position data are already over 5MB compressed. So small deviations are worth the tradeoff.</p><p>The result of all this wrangling is <a href="https://spcotter.github.io/visualizing-pittsburgh-transit/canvas.html"><strong>a pretty smooth animation of all weekday trips</strong></a>. At peak times, there are over 430 active trips throughout the county. That’s a lot of of buses. So the next time your bus is late, remember there are hundreds of other buses out there trying to get thousands of other people where they need to go. Designing and managing such a system can’t be easy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*HoT3RwJ06zLKB0odr4ojog.gif" /><figcaption>All trips at around 6:00 AM.</figcaption></figure><p>Leaflet takes care of nearly all the details of zooming in to the animation. I only have to adjust the size of circles so they’re not huge at high zoom levels. I also added controls to allow pausing and adjusting the speed.</p><p>Getting closer makes it easier to see how much is going on downtown during rush hour. Many of those 6,500 trips start or end downtown. Notice how quickly buses get in and out on the busways compared to mixed traffic like Liberty Avenue or 5th Avenue.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*0uKeSIuVWB3_IJT_tGzs8w.gif" /><figcaption>Buses circulating downtown at around 6:00 AM on a weekday.</figcaption></figure><p>Rush hour in Oakland makes it clear why Port Authority is considering bus rapid transit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*Rd3ojdiEkr6jFnL6TSS3zA.gif" /><figcaption>Buses running through Oakland at around 4:30 PM on a weekday.</figcaption></figure><p>If you’re interested in the details of data manipulation and animation, all the code is <a href="https://github.com/spcotter/visualizing-pittsburgh-transit">available on GitHub</a>. Refer to the README file for an overview of the process. The code could be modified for another transit system’s GTFS data.</p><p>To improve the animation, the zoom behavior could adjust the frame rate and interpolation. Real-time data could also be used to show actual rather than scheduled trips. But Port Authority’s API restrictions and rate limits make it difficult to gather all the data at sufficient intervals.</p><p>Have comments or other ideas for improvement? Leave a response or fork the repository!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ab05664b8b3" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>