Thanks for the find, Fergal. A few interesting points:
1. The winners in this article only came second on the kaggle leaderboard
https://www.kaggle.com/c/seizure-prediction/leaderboard
Perhaps the top team did not share their code?
2. The winners describe their methodology as a weighted combination of
three models (Generalized Linear Model, Random Forest and Support Vector
Machines). Each model used a different method of data reduction such as
hand-picked FFT power bands, standard brain wave FFT bands and feature
detection. They tried many combinations to see what gave the best kaggle
score.
https://github.com/drewabbot/kaggle-seizure-prediction/blob/master/report.md
3. This is very different to the pure NuPIC attempt where we tried to
learn from clean time histories and look for anomaly likelihood in each
test signal. Basically we ran out of time having started late and never
got to process all 5 dogs and 2 human patients to even see if this
approach got close. Based on the differences in time history "noise" in
the dog and patient data I saw, more data reduction work is needed.
Perhaps a spectrogram time history is what we could feed NuPIC?
If anyone is interested in progressing this we can still submit results
to see if we do better than the "winners". Let me know if you are
interested in helping on this work.
Andrew Currie
Sydney
On 12/12/14 00:09, Fergal Byrne wrote:
Nice article on the competition (which a NuPIC team participated in):
http://www.npr.org/blogs/health/2014/12/10/369654830/a-crowd-of-scientists-finds-a-better-way-to-predict-epileptic-seizures
--
Fergal Byrne, Brenter IT
http://inbits.com <http://inbits.com/> - Better Living through
Thoughtful Technology
http://ie.linkedin.com/in/fergbyrne/ - https://github.com/fergalbyrne
Founder of Clortex: HTM in Clojure -
https://github.com/nupic-community/clortex
Author, Real Machine Intelligence with Clortex and NuPIC
Read for free or buy the book at https://leanpub.com/realsmartmachines
Speaking on Clortex and HTM/CLA at euroClojure Krakow, June 2014:
http://euroclojure.com/2014/
and at LambdaJam Chicago, July 2014: http://www.lambdajam.com
e:[email protected]
<mailto:[email protected]> t:+353 83 4214179
Join the quest for Machine Intelligence at http://numenta.org
Formerly of Adnet [email protected] <mailto:[email protected]>
http://www.adnet.ie <http://www.adnet.ie/>
