Kaggle UPenn and Mayo Clinic's Seizure Detection Challenge

This repository contains documentation and code for the second place submission by Eben Olson and Damian Mingle.

http://www.kaggle.com/c/seizure-detection

##Hardware / OS platform used

• EC2 m3.2xlarge instances (8 vCPU, 30GB RAM)
• Core i5-2400 quad core @ 3.10GHz, 16GB RAM
• Ubuntu linux 12.04

##Dependencies

• Python 2.7
• IPython 2.1.0
• Theano 0.6.0
• scikit_learn-0.14.1
• numpy-1.8.1
• scipy-0.14.0
• joblib-0.8.3-r1 (only needed for multi-core batch processing)

##Steps to train the model and obtain a submission

• Obtain the competition data and uncompress it into the 'data/clips' directory of the project.

data/
  clips/
    Dog_1/
      Dog_1_ictal_segment_1.mat
      Dog_1_ictal_segment_2.mat
      ...
      Dog_1_interictal_segment_1.mat
      Dog_1_interictal_segment_2.mat
      ...
      Dog_1_test_segment_1.mat
      Dog_1_test_segment_2.mat
      ...

    Dog_2/
    ...

• Run IPython from inside the project directory with pylab mode enabled

ipython --pylab

• Open the notebook "Data Preprocessing.ipnyb" in your browser and execute all cells to prepare data for training.

• Open the notebook "Train Classifiers and Predict.ipynb" and execute all cells to train models and save predictions to the "output" directory.

• Open the notebook "Postprocessing.ipynb" and execute all cells to average the model predictions and produce "submission.csv" in the project root directory.

View notebooks online at NBViewer

• Data Preprocessing
• Train Classifiers and Predict
• Postprocessing

Project report

Available at https://github.com/ebenolson/seizure-detection/raw/master/report.pdf