HAR-Pipeline

Prerequisites

• Python 2.7
  • Numpy
  • Tensorflow
  • Pandas

File descrtiption:

• training.py. Training both the Convolutional Neural Network and the Viterbi transition matrix. Use data located in DATA/TRAINING.

• testing.py. Runs the data located in DATA/TESTING through the Convolutional Neural Network and the Viterbi model. Saves the predictions and statistics in the folder RESULTS/

• predicting.py. Predicting unlabelled data. Saves the prediction in RESULTS/RESULT_PREDICTING.py

• cnn.py. Covolutional Neural Network, used both for training and testing. Configuration is done via the TRAINING_VARIABLES.py file.

• viterbi.py. Viterbi, used both for training and testing.

• data.py. Loads the different data sets. Splits the raw signal into windows used in the network.

• TRAINING_VARIABLES. All variables used for training and testing.

Training and testing folders/files

The data from each subject must be placed in a spesific folder under DATA/TRAINING/ or DATA/TESTING.
Example: DATA/TRAINING/A01

Inside that folder, the three files must contain the name of the sensor ("BACK", "THIGH" and "LAB"). The name must be separated by two underscores.
Example:

• DATA/TRAINING/A01/01A_Axivity_BACK_Back.csv
• DATA/TRAINING/A01/01A_Axivity_THIGH_Right.csv
• DATA/TRAINING/A01/01A_GoPro_LAB_All.csv

Predicting folders/files

Same structure as training and testing (explained over), but without annotation file (01A_GoPro_LAB_All.csv).

Data structure

The training and testing data need to be in a spesific format.
Sensor format:

-1.0156,-0.079657,-0.0015319
-1.0303,-0.079044,-0.016544
-1.0312,-0.09375,-0.03125
-1.0156,-0.078125,-0.03125
-1.015,-0.078125,-0.03125
-1.0147,-0.078125,-0.032169
...

Label format:

17
17
6
6
6
6
...

Training

In terminal:

python training.py

Testing

In terminal:

python testing.py