MobiRNN-EMDL17

Note that this repo is for reproducing the results reported in the MobiRNN paper(v1). For the latest MobiRNN, please see this repo.

Usage

Data folder has a zip file which needs to be unzipped and placed on the sdcard of Android smartphone.

Model folder contains Python code for LSTM model on smartphone sensor data using default configuration (2 layer, 32 hidden units). Other layer and hidden units configurations can be found at different tags in this repo. For example, 3 layer 128 units, see tag v3.128

Android implementation of MobiRNN is in the Code folder. Cuda-based version can be found at the cuda branch of this repo, where util branch takes GPU utilization into account.

Acknowledgement

The original model source code is modified from guillaume-chevalier