For the last year I've been working hard developing a couple of projects which I think are quite interesting, so I'd like to share them with you. They are not using TGC products so I'm posting it here rather than in WIP/program announcements.
GeoTweetSearch
First is a project called GeoTweetSearch which analyses Twitter data streams. You create a 'search stream' which looks at a certain geographic area or tweets containing a set of keywords/hashtags. Then all tweets which match this criteria are analysed; we process up to 180,000 tweets per hour per stream (this limit is imposed by Twitter).
The project is written in Python and uses a MongoDB back end and web based (lots of javascript) front end.
We geocode tweets by looking at geotagged tweets from smart phones and other devices which attach a real world coordinate to tweets; but this only accounts for about 2% of tweets. So we also look at the location field of Twitter user profiles and infer their location from that. We are able to get about 20-40% geocoding success rate with reasonable accuracy.
We also generate influence data by looking at users' followers and we say for example, if 50% of a user's followers are based in London then the user has a high influence on London. We also do this at a location level and are able to say things like for example 'users in London have a high influence on New York'.
Link to the project:
www.cs.bham.ac.uk/research/projects/nsl/geotweets
Note: the MongoDB instance seems quite unstable at the moment so the website may go down from time to time. So here are some screenshots:
http://mikenetapicouk.fatcow.com/personal/geo_tweet_search_search_stream.png
http://mikenetapicouk.fatcow.com/personal/geo_tweet_search_search_stream2.png
http://mikenetapicouk.fatcow.com/personal/geo_tweet_search_london_live.png
http://mikenetapicouk.fatcow.com/personal/geo_tweet_search_user_view.png
CUDA Disease Propagation Simulation
This I think is really cool! It is written in C++ (with OpenGL for the display) and CUDA and runs a high performance disease propagation simulation, with most of the computation running directly on the GPU (CUDA is essentially C with the ability to run code directly on the GPU). The simulation is a good way of benchmarking your GPU; as you increase the number of actors, the amount of computation increases and this will impact the frame rate.
This is open source and you can find the repository here:
https://github.com/watfordxp/Disease-Propagation-Simulation
If you have a graphics card which is CUDA enabled, you can download an executable for Windows and run the simulation from here:
http://mikenetapicouk.fatcow.com/personal/disease_propagation.zip
I noticed that Windows 8 blocks it for some reason, but you can run it from command line fine. Also if you run from command line you can specify an integer argument to specify screen width and height in pixels so: "disease_propagation.exe 500" will set screen width and height to 500 pixels.
Here is a screenshot:
and some more:
http://mikenetapicouk.fatcow.com/personal/alot_of_actors_finished_1.png
http://mikenetapicouk.fatcow.com/personal/alot_of_actors_finished_2.png
http://mikenetapicouk.fatcow.com/personal/just_started.png
http://mikenetapicouk.fatcow.com/personal/infection_radius.png
