View the corresponding article on my github pages site obrhubr.github.io.
Use the awesome Sensor Logger App that is available for both Android and iOS. Choose accelerometer, gyrometer and GPS data for logging and press start. Then hit the slopes and worry about data science later.
Create a data directory here, and add train and test to it.
Then, export your data from the Data Logger App in csv format, both training and test data.
Put the zip files for each in their corresponding directories, renaming the files to data.zip.
First, open data_wrangle.ipynb and run the scripts to extract the data and reshape it. To label it, use data_label.ipynb, making sure to correctly edit the json object containing the labels with your own data. Then use data_prepare.ipynb to create an actual dataset.
You should now have multiple files called train_100.csv and train_quat_400.csv in your data folders. To train your model on them, use the notebook model_train.ipynb, making sure to load the right dataset with the right values for channel and datapoints (6 channels if you use 6 features from the sensors, 400 datapoints if you choose train_400.csv for example).
To get the predictions, use model_predict.ipynb.
All other notebooks were experiments used to optimise the model.