Readme.md

ML estimator model for insulin suggestions plugin

Setup

Required software:

• python 3.6
• pip3

Install dependencies

pip3 install -r requirements.txt

Prepare the data for training

Export data from the app

From the app settings, click on "Export data" and authenticate to allow the app to save your glucose data for ml training.

Import using adb

Enable usb debugging on your device, plug it in and then run on your pc

bash import.sh data

Import manually

Copy the files from the Documents directory inside the device memory to the 'data' directory.

Setup the tests

Fill the data/test_*.csv files with data with desired results (pick some from the train files).

Train the model and get the results

The app uses some json files to look at references of the model results instead of shipping the real model itself in order to allow lower-end devices to use this feature too.

All you need to do is execute the python script:

python3 estimator.py

Install the plugin

Once the training has been completed, the plugin will be available at the following path from the repository root: ml/export/plugin.zip.

Send this file to the device, open the app settings, click on "Suggestions plugin", select the zip file and wait until it's installed.

To update the plugin, repeat this procedure.

Disable tracking git changes of sensitive files

Sensitive file changes tracking should be disabled by running the setup.sh file found in the root of this repository, or by updating the git index manually:

 git update-index --skip-worktree ml/data/*
 git update-index --skip-worktree ml/export/*