Tube Sign – Service Information image generation service

Part of a series of posts to cover some small projects that I did whilst not being able to work. They cover things from the role of familiar strangers on the internet and anti-social networks, through to meteorological hacks, funny memes to twitter bots. This post is about a funny meme image generation service.

Sometimes I surf the internet for funny pictures. Although the ones with cats I have a healthy distrust for – there was one class of amusing image which caught my eye. Those funny or inspirational London Underground passenger informations signs. I was seeing these every week and thought… “I could do that”. So I did, created tubesign.herokuapp.com and a few other people found it funny. At one point there was about 50 people visiting at any one time and when I put the statistics on there was 13,000 views on the second day with an image being created one every second. At time of writing it has had over 50,000 views.

from http://weknowmemes.com/wp-content/uploads/2012/09/apple-maps-london-tube-sign.jpg

An actual real life TfL service information Sign! from http://weknowmemes.com/wp-content/uploads/2012/09/apple-maps-london-tube-sign.jpg

How I did it.

First of all I looked into fonts – I wanted to get a good handwriting font which would look as if someone had used a marker on a white board. Google fonts delivered, and I chose Reenie Beany.

It uses Sinatra, Ruby and Rmagick and is hosted on the Heroku platform – even at it’s busiest it was able to cope on the free tier. It doesnt use any database. It caches requests for images though.

I use a bit of random number generation to change the angle the text is written at, and change the indent a bit.

The code for Tube Sign is on github  but give it a go firsttubesign.herokuapp.com

Viral & coverage

I posted this on facebook and my friends gave it a go, with some hilarious images being created, and then it spread to twitter, where more and more people found it. Then blogs, mainly London based blogs found it.

first image used – the source for this I could not found, and so the use of this image was discontinued

Someone said that the original image was someone’s copyright, so I changed it to a CC-By-SA image by Flickr user Lrosa, which also meant that all images created were under the same licence.

Creative Commons by Share Alike, Attribution image which is the image being used on the application. Image from Flickr, Lrosa, http://www.flickr.com/photos/lrosa/1138285047/

The main media outlets that covered it were: BBC America, ITV, The Londonist, The Atlantic Cities, The Guardian, The Next Web and the B3ta.com newsletter (very proud of that one).

There was about 50 people visiting at any one time and when I put the statistics on there was 13,000 views on the second day with an image being created one every second. At time of writing it has had over 50,000 views. Now the traffic is in the hundreds, with number of people visiting right now enough to be counted on one hand.

Future

  • Live preview
  • Better font rendering – defocus
  • Add range of images for different places (Bombay signs, Leeds Metro signs etc)
  • Store images, allow voting, create gallery

Rain Prediction for the immediate future using Met Office DataPoint – rain graph

Part of a series of posts to cover some small projects that I did whilst not being able to work. They cover things from the role of familiar strangers on the internet and anti-social networks, through to meteorological hacks, funny memes to twitter bots. This post is about a meteorological hack.

The Met Office in the UK have in this last year published an API for their range of services – all part of the open data movement I think.

DataPoint is a way of accessing freely available Met Office data feeds in a format that is suitable for application developers. It is aimed at professionals, the scientific community and student or amateur developers, in fact anyone looking to re-use Met Office data within their own innovative applications.

The year before this, in Denver, USA, I was shown by a couple of awesome mapping and weather geeks a mobile app that showed when it was going to rain, and more importantly when it wasn’t going to rain in very high temporal resolution. You can use the app and know whether to get a swift half and then leave to get the bus, or whether to stay in for an hour until the showers end. It was very detailed and highly useful. This app Dark Sky App

was freaking awesome. And I want here in the UK, so when the Met Office announced their API I was interested.

You cannot do what DarkSkyApp does with the Met Office DataPoint API though – what you can do is do some interpolations though. The API for precipitation forecasts only gives access to a 3 hourly map tile.

http://www.metoffice.gov.uk/datapoint/product/precipitation-forecast-map-layer

Although further poking around shows that they do have an undocumented 1 hourly image.

Screenshot - 061213 - 16:05:15

These map tiles then could be used. http://rain-graph.herokuapp.com  is the resulting application with the code here: https://github.com/timwaters/rain_graph

It’s a Ruby Sinatra application which for a location, grabs the precipitation tile for a defined location for each hour from now. It looks at the pixel value for the given location and determines the amount predicted. It shows when the heaviest rain is predicted and when it would stop. Interpolation is given by the graph engine itself – no fancy meteorological modelling is done (at this stage). It uses Chunky_png to get the pixel values.

687474703a2f2f692e696d6775722e636f6d2f4343306f5868772e706e67

All requests are cached to avoid hitting the MetOffice API and because an image won’t change for an hour. Additionally it uses another API method to get a human readable upcoming forecast text for that location, and displays it under the graph. Contrary to popular global belief it’s not always raining in the UK, and so most of the time you will never see a a graph showing something!

Considerations:

Pixels to Lat Lon:
Since a lat/lon location is quite specific, it could map to one pixel in a tile, and that pixel could have a lower or higher value than the ones surrounding it. I could use a kernel average – do a 6×6 pass over the pixel and get the average value. But since there are tiles are lower zoom levels, by zooming out, the spatial extent of the pixel would equal that larger area – it would do the work for us.

Interpolation between forecasts:
It wasn’t clear if the forecast images showed the predicted situation over the whole hour, or whether it showed the situation at that moment. Should a user look at an animation to see how rain cloud moves across from A->B and guess that in between that there would be rain, or should they think that that there would be no rain if there is no rain shown?

User Interface:
It looks a bit bland – we should show the image tiles underneath  – perhaps shown when hovering over a point.

Accuracy:
I haven’t tested the accuracy of this.

Location hard coding:
The text forecasts are hardcoded to a set number of regions, but we could do a closest point and get the correct forecast for the given lat and lon.

Use Yr.No API

Yr.no has detailed hour by hour forecasts API for a place giving the amount of precipitation.
http://www.yr.no/place/United_Kingdom/England/Leeds/hour_by_hour_detailed.html

<time from="2013-12-06T19:00:00" to="2013-12-06T20:00:00">
<!-- Valid from 2013-12-06T19:00:00 to 2013-12-06T20:00:00 -->
<symbol number="3" name="Partly cloudy" var="mf/03n.11" />
<precipitation value="0" /><!-- Valid at 2013-12-06T19:00:00 -->
<windDirection deg="294.2" code="WNW" name="West-northwest" />
<windSpeed mps="4.3" name="Gentle breeze" />
<temperature unit="celsius" value="1" />
<pressure unit="hPa" value="1004.9" />
</time>

Markov Chains, Twitter and Radical Texts

The next few posts will cover some pet projects that I did whilst not being able to work due to recent civic duty.  They cover things from the role of familiar strangers on the internet and anti-social networks, through to meteorological hacks, funny memes to twitter bots. The first in this series is about what happens when you use markov chains and radical texts with twitter.

Detournement is a technique now considered to the father of remixes or mashups, but with a satirical political nature. Have a look at the wikipedia entry for detournement if you want to know more about it. Basically you do something to something which twists or re routes it so that it makes new meanings. It was the Situationists, led by Debord who really adopted and ran with this as a practice.

guy_debord1

Debord would often frequently plagiarise other radical texts in his own work. (The Situationists were also the ones behind original notion of psychogeography – something that you may have caught me talking about before.)

So what would happen if we could detourn, or mashup, or plagiarise Debord’s own writings? And how about if we could publish it periodically, and how about if we had a 140 character limit? Yeah so this is my experiments with these ideas.

Bruna Rizzi; it is from this disastrous exaggeration. The peasant class could not recognize the practical change of products

The proletariat is objectively reinforced by the progressive disappearance of the globe as the bureaucracy can

Markov chains basically work like take a couple of sentences: “A lazy dog likes cheese” and “My house likes to be clean” then look at groups of two or three words together. Then if one of these groups share the same word (“likes”), make a new sentence using that word to chain together. “My house likes cheese” or “A lazy dog likes to be clean”. Markov chains result in sentences that look human readable. The more sentences you feed the population sample, the better or more varied the same of generated sentences.

Some radical texts are complete nonsense and really hard to read, so perhaps applying Markov chains to them can help reveal what truths the obscure language hide.

@markov = MarkyMarkov::TemporaryDictionary.new
@markov.parse_file "debord.txt"
raw_text = @markov.generate_23_words

My solution uses Ruby, the Twitter gem and the marky_markov gem.

https://github.com/timwaters/rattoo  is the work in progress twitter bot – it works currently on Heroku using the scheduler to periodically tweet a sentence, see if any other users have asked it questions and reply back to them.