This is an Apple Watch Application(Use Swift) designed for testing the energy consumption of different models when performing the human activation detection task. Currently, the code for the generation of synthetic data --> data preprocessing --> model inference --> and displaying prediction results for this functionality has been thoroughly tested.
Under the premise of controlling the adjustable parameters, the performance of five models (CNN, TensorFlow, GRU, LSTM, TensorFlow+CNN) on the same data set was tested.
Combining several indicators such as accuracy and using time (Testing Results.xlsx), the GRU model has the best score.
coreModel format is the model format that swift accept. The file tf_to_coreModel.ipynp is used to convert tensorflow format to coreModle format.
The document includes the original dataset we used WISDM. It also includes the preprocessed data: WISDM_x.csv(Attributes), WISDM_y.csv(Predict Target).
- mpSensors: a mobile app for the activity recognition experiments documented above. It is developed in Kotlin and supports Android and iOS.