Skip to content

Latest commit

 

History

History
64 lines (46 loc) · 2.19 KB

readme.md

File metadata and controls

64 lines (46 loc) · 2.19 KB
Raw

Crowd-ML - Android

Add Firebase to your app

Note: Your app will not build until this is finished

  1. Download Android Studio 1.5 or higher
  2. Install Android SDK Manager
  3. Download Android Studio project and retrieve the package name
  4. Paste the training data feature and label files (found in the data folder) and paste them into Crowd-ML/client/android/Crowd-ML/app/src/main/assets/
  5. From the Firebase console, go to your Firebase project and click 'Add Firebase to your Android app' and follow the steps
  6. Downloaded and copy the google-services.json file into your app's module folder (typically .../app/)

Setup

Setup the following environment variables for your environment:

# Controls output directory in APK and CPU type for Bazel builds.
# Ex: `armeabi-v7a` or `arm64-v8a`
export CROWDML__CPU_TYPE="<cpu type>"

# Path to TensorFlow source
# Ex: '/home/user/workspace/tensorflow'
export CROWDML__PATH_TO_TF="<path to tensorflow>"

NOTE: ANDROID_NDK may also need to be set to point to the given ndk

Add the SDK

Most dependencies are universal between Android projects and are included in the download, however if you wish to add more features, you may need to add additional Firebase dependencies. To do this, in your module Gradle file (in the app/build.gradle) file, add the apply plugin line at the bottom of the file to enable the Gradle plugin:

   apply plugin: 'com.android.application'
   
   android {
   // ...
   }
   
   dependencies {
   // ...
   compile 'com.google.firebase:firebase-core:9.0.2'
   //<Additional dependencies go here>
   }

A more detailed instruction and list of additional SDKs can be found here you may wish to use:

https://firebase.google.com/docs/android/setup

Configuring Android app

When the app begins, it will download the parameters from the firebase database, which have been uploaded by the server. These consist of:

  • regularization constant
  • noise variance
  • noise distribution
  • number of classes
  • loss function
  • label and feature source files (these must be located in the assets folder)
  • featureSize
  • number of training samples
  • gradient batch size.