This repository contains the code that was used by the team Spreedictors - Wolfgang Kopp and Gregor Lichtner - to participate in the LDopa subchallenges of the Parkinson's disease DREAM challenge.
To run the scripts you need the following software requirements:
- Install Anaconda2-4.4.0.
The requirements that were used to train and evaluate the models can be loaded via
conda create --name <env> --file requirements.txt
You need to set the environment variable PARKINSON_DREAM_LDOPA_DATA to point to
the directory where the dataset should be stored. For instance,
on Linux use
export PARKINSON_DREAM_LDOPA_DATA=/path/to/data/
Furthermore, you might need to set KERAS_BACKEND to utilize tensorflow
rather than e.g. theano according to
export KERAS_BACKEND=tensorflow
or by running python with
KERAS_BACKEND=tensorflow python <script.py>
To train the models we invoked
cd <repo_root>/code
# Subchallenge 2.1: tremorScore
# variant 1
python run_all.py -df fh_0.5-tre-all -mf metatime_deep_conv_v2 --allaug
# variant 2
python run_all.py -df raw-tre-all -mf metatime_deep_conv_v2 --allaug
# Subchallenge 2.2: dyskinesia
# variant 1
python run_all.py -df raw-dys-all -mf metatime_conv2l_70_200_10_50_30_20_10 --allaug
# variant 2
python run_all.py -df raw-dys-all -mf metatime_deep_conv_v2 --allaug
# Subchallenge 2.3: bradykinesia
# variant 1
python run_all.py -df raw-bra-all -mf metatime_deep_conv_v2 --allaug_v2
# variant 2
python run_all.py -df fh_0.5-bra-all -mf metatime_deep_conv_v2 --allaug
These commands will automatically download and preprocess the LDopa dataset provided for the subchallenges 2.1-2.3.
Finally, the feature predictions were generated and submitted to the challenge submission queue with
python featurizer.py tre1 --gen --submit
python featurizer.py tre2 --gen --submit
python featurizer.py bra1 --gen --submit
python featurizer.py bra2 --gen --submit
python featurizer.py dys1 --gen --submit
python featurizer.py dys2 --gen --submit