PDBChallenge

Parkinson's Disease Digital Biomarker Dream Challenge

We have the following directory structure:

project
│   Walking_Activity_Train.csv
│   Walking_Activity_Test.csv
|   Walking_Activity_Supplemental.csv
│
└───Train
│   │
│   └───deviceMotion_walking_outbound
│       │   123
│       │   234
│       │   ...(the downloaded directory structure for deviceMotion_walking_outbound column)
│   │
│   └───deviceMotion_walking_outbound
│   └───deviceMotion_walking_rest
│   └───deviceMotion_walking_return
│   └───accel_walking_outbound
│   └───accel_walking_rest
│   └───acc_walking_return
│   
└───Test
|   │   
─── Supplemental
|   |

For every folder such as deviceMotion_walking_outbound, we use a command python common.py --x_y_z_norm Y --function_no 0& to calculate the the meanY feature for the files.

--x_y_z_norm is the argument for the type of feature, X,Y,Z or norm.. --function_no is the function no from 0 to 11.

Similarly, there are 12 different functions that we have calculated,

• 0 meanX mean of the X acceleration series
• 1 sdX standard deviation of the X acceleration series
• 2 skewX skewness of the X acceleration series
• 3 kurtosisX kurtosis of the X acceleration series
• 4 q1X first quartile of the X acceleration series
• 5 q3X third quartile of the X acceleration series
• 6 iqrX interquartile range of the X acceleration series
• 7 ptpX range of the X acceleration series
• 8 autocorrX autocorrelation (lag = 1) of the X acceleration series
• 9 zcrX zero-crossing rate of the X acceleration series
• 10 dfaX scaling exponent of the detrended fluctuation analysis of the X acceleration series
• 11 variationX coefficient of variation of the X acceleration series

We then calculate them for each of the 4 types of features, and merge them using the merge code.

Libraries

• scikit-learn

• nolds for dfa calculation( https://pypi.python.org/pypi/nolds )

• numpy, scipy

• Python Version 3.6