Machine Learning Apps with Swift as Backend & Python ML Models

There has been demand of apps consuming Machine Learning Models to predict results. Apple & Android have been evolving their base day by day to provide better experience to Developers & End Users both. There has been several challanges in consumption of Trained Models in apps —

• Hardware & OS compatibility
• Duplicate development efforts in Apps Development to consume same trained models
• Separate Trained Models for different OS — High maintenance cost

App Developers world has been evolving day by day with improved tech stack. In this article — I have come up with one of unique combination for better results.

Swift as Backend — using Vapor

• Swift 5.5
• Vapor 4
• Python3

AWS

• Ubuntu 20.04
• t2.large instance

Machine Learning (trained) Model — SAV (python)

Native / Cross Platform Apps — consuming the APIs — Hope we already know this ;-)

Lets get started! XCode Vapor Project — On Mac Machine

Create a Swift Package Project using Vapor. Install Vapor on your Mac Machine. Follow below documentation —

Vapor: Install → macOS

Install Python3 on Mac machine. Mac machines generally has two python versions -

python — v2.7
python3 — v3.8 (or any other version of python v3)

Execute below command on Mac Machine to update to latest version of Python

brew install python

This updates python3 mapping of usr/bin to brew installed python. Add ENV variable in your .bash_profile / .zshrc / .profile / .zprofile

export PYTHON_LIBRARY=/usr/bin/python3

In terminal — source your profile file where you added ENV variable

source ~/.zshrc

Machine Learning Model — SAV (python)

Check your python sample code — accessing the same ML Model, install the required “import” libraries. Generally, below are the tools needed by PythonML Models

joblib

pip3 install joblib

pandas

pip3 install pandas

scikit-learn

pip3 install -U scikit-learn scipy matplotlib

XCode Vapor Project

Add PythonKit to package file dependencies

.package(url: “https://github.com/pvieito/PythonKit.git", .branch(“master”))

Add python models — SAV files — in Project Sources > App

Add SAV models in resources section of package.swift file — path is relative to App Folder inside sources

resources: [.copy(“Python/HeartRiskModel.sav”)],

Now, in Swift File follow the below —

import PythonKit
class HeartRiskPythonModel {
/// URL of model assuming it was installed in the same bundle as this class
    class var urlOfModelInThisBundle : String {
        let _ = Bundle(for: self)
        return Bundle.module.path(forResource: “HeartRiskModel”, ofType:”sav”)!
    }
}

In Routes file, or the target file, where you want to use ML Model — load model classes as class level variables, using below

let joblib = Python.import(“joblib”)
let heartModelPath = HeartRiskPythonModel.urlOfModelInThisBundle
let heartRiskModel = joblib.load(heartModelPath)

Now, in your target method, use below —

let user_data = [[body.age, body.sex, body.cp, body.trestbps, body.chol, body.fbs, body.restecg, body.thalach, body.exang, body.oldpeak, body.slope, body.ca, body.thal]]
let predection = heartRiskModel.predict(user_data)
let predection_prob = heartRiskModel.predict_proba(user_data)

You can obviously achieve the same Using CoreML library of Apple, which is straight forward BUT that comes up with other restrictions-

• You need to set-up Mac Instance on AWS — which is expensive
• You always need to map Python Models into CoreML models

Lets GO LIVE! — On AWS, login using credentials and SetUp new EC2 OR user your existing one — make sure you are having t2.large instance as minimum for this.

Using Root User — Install Swift following below commands — Swift Version Download link in step 4 of below — can be found at https://www.swift.org/download/

sudo apt update
sudo apt upgrade
sudo apt install binutils git gnupg2 libc6-dev libcurl4 libedit2 libgcc-9-dev libpython2.7 libsqlite3-0 libstdc++-9-dev libxml2 libz3-dev pkg-config tzdata zlib1g-dev
wget https://swift.org/builds/swift-5.3.3-release/ubuntu2004/swift-5.3.3-RELEASE/swift-5.3.3-RELEASE-ubuntu20.04.tar.gz
tar xzf swift-5.3.3-RELEASE-ubuntu20.04.tar.gz
sudo mv swift-5.3.3-RELEASE-ubuntu20.04 /usr/share/swift
echo "export PATH=/usr/share/swift/usr/bin:$PATH" >> ~/.bashrc
source  ~/.bashrc
swift --version

Install Vapor using below documentation — (I preferred toolbox approach)

Ubuntu - Vapor Docs

Install Python & Required dependencies on AWS too, using the same commands as stated above.

Git Clone your Vapor Porject — move to code directory and run project using below — (I did this using root user only)

vapor run serve — hostname 0.0.0.0 — port 80

All SET!! You can now consume the API on both iOS & Android Apps.