Table of Contents

• Integrating R Modeling Into the Esri Ecosystem
  • Live Links
  • Workflow
  • Key Files
    • Client Side
    • Server Side
• Frontend Creation
  • Creating a Blank Web AppBuilder App
  • Run the Frontend App
  • Add Layers from ArcGIS Online
  • Add a Custom Widget
• Backend Server Creation
  • Linking Frontend to Server
  • Deploying Remote Server
    • Running the R Model
    • Creating an Friendly API

Integrating R Modeling Into the Esri Ecosystem

This repository provides an example of how a web-based Esri frontend can be combined with the R programming language. This allows combining spatial data and analytics from the Esri ecosystem with the statistical programming and modeling capabilities of R.

Live Links

To get an idea of what's possible, take a look at the live demo frontend, which is an Esri-based map that calls an R-based backend to run forecasts that incorporate spatial data:

Live Demo: https://retailscientifics.github.io/Esri-R-Integration/

We've also created several R Markdown Notebooks that walk through the process of training a model and testing it locally:

R Notebook: Building and Training Models: https://retailscientifics.github.io/Esri-R-Integration/Server/training.html

R Notebook: Local Testing and Prediction: https://retailscientifics.github.io/Esri-R-Integration/Server/standalone_test.html

Workflow

The workflow demonstrated in this repository goes something like this:

• Build a model locally within R
  • Generally predicts a response variable from a number of regressors
  • Regressors can include spatial data sourced from Esri shapefiles
• Test model locally by running predictions on mocked-up input data
• Productionalize your code by bundling the data staging and prediction steps into standalone functions
  • Then use the plumber library to serve this function as an API endpoint on a remote server
• Create a web map frontend with a form to supply new input data/regressors (based on mockup from previous step)
• Write code for your form at API endpoint to return new predictions (and optionally other data as well)

Key Files

Client Side

On the client side, the key production files for the web-based frontend will be:

• config.json
  • Shows how to add a custom widget to WebappBuilder
• index.html
  • Includes modifications for pulling in external javascript libraries such as jQuery
• widgets/RIntegration/Widget.html
  • The HTML template for the custom widget, including an input form
• widgets/RIntegration/Widget.js
  • The code/logic for pushing the form to a server-side R API to obtain a prediction

Nearly everything else is proved by Esri's WebappBuilder framework.

Server Side

And on the server side, the key production files are

• Server/api.R
  • A "productionalized" version of model code, which describes an API endpoint that receives new data, adjoins new variables (including spatial data from a shapefile), and runs the model to obtain a prediction
• Server/esri_demo.R
  • The main production file that is run. Sets up the server, sources all of the appropriate libraries and static files, and serves up the endpoint defined in api.R

To help demonstrate how these files came to be and how they function, the following notebooks have also been included:

• Server/training.Rmd
  • Shows how the pre-production step of building and testing a model can be carried out.
• Server/standalone_test.Rmd
  • Step-by-step mockups of how the server-side R code will receive new data and run the model on it.

This repository can be downloaded and used as-is; the following describes how to reproduce this repository from scratch.

Frontend Creation

Creating a Blank Web AppBuilder App

Note: you can skip this step by simply downloading this repository

• If you don't have it already, install node.js
• Download the latest version of Web AppBuilder for ArcGIS
• Unzip and open a terminal in the resulting directory
• cd server
• npm install
• npm run start
• Open a browser to the page that is served up - in this case, localhost:3346.
• Enter the URL for your ArcGIS Online Portal (in our case, http://scientifics.maps.arcgis.com/)
• It will now ask for an App ID - click "Help" and follow the instructions laid out there. For the URL, use http://localhost:3346/webappbuilder.
• When it is created, go to "Settings -> App Registration" to register an ID, using the same URL as the redirect.
  • Alternatively: Back on your localhost page, create a default 2D web map and save it. Navigate back to the main page, find the map you just made, and save it.
  • Unzip it to a directory.

Run the Frontend App

• Option 1: Simply copy all of these files to a server somewhere and host them online to access.
• Option 2: Start a local https server:
  • First generate a self-signed SSL certificate to enable HTTPS: openssl req -new -x509 -keyout server.pem -out server.pem -days 365 -nodes
  • Run python3 server.py (note: this does require python 3) and navigate to https://localhost:8000/
• You may get an "unsafe connection", this can be bypassed for testing purposes. You should now be able to see a basic webmap:

Add Layers from ArcGIS Online

To do!

Add a Custom Widget

• Install the widget generator with npm install -g yo generator-esri-appbuilder-js
• Make a new directory for your project and run yo esri-appbuilder-js to generate some initial files and set some settings.
• Run yo esri-appbuilder-js:widget to create your actual widget in the ./widgets directory. This will provide the basic scaffolding around which a full widget can be built.
• Copy your new widget back into your app's widget directory. For us, this will be under widgets/RIntegration

Backend Server Creation

• Install R, by either following the instructions at one of the websites or
  • On unix operating systems, running something equivalent to sudo apt-get install r-base suffices.
• (Optional) Install RStudio.
• Install the plumber package:
  • Launch R from the command line by typing R (or launch RStudio), and at the prompt, install plumber package by typing install.packages('plumber').
• Now you can set up a test endpoint. Here is an example:

   # In a file such as 'test.R'
  
   #* @get /sum
   function(a, b) {
   	return(a + b)
   }
  
   # In a separate file such as 'server.R' or in the R command prompt
   library(plumber)
   r <- plumb('test.R') 	# The file made above
   r$run(port=8000) 	# Any free port

  After entering these commands or sourcing server.R, you can now visit something like http://localhost:8001/sum?a=1&b=2 in your browser to test the results.

Linking Frontend to Server

Note: setup is simplest when both the frontend web map and the back end server are hosted on the same machine and accessible via the same domain; otherwise, you may run into issues related to Cross-Origin Resource Sharing (CORS).

• Set your working directory to Server and source esri_demo.R, which starts plumber and begins serving the endpoint defined in api.R.
  • In Unix, you can make the file executable with chmod +x esri_demo.R and then type ./esri_demo.R to launch.
• Check that the following link works in your browser: http://localhost:8001/runmodel?inputDataframe={"Latitude":38.98730482188786,"Longitude":-114.84300214791408,"LocationName":"asdsa","LocationSquareFootage":"123","locationLatitudeLongitude":"38.98730,+-114.84300","PopulationDensity":"low","PropBoomers":"average","NeighborsToUse":"4"}&apikey=some_secret_key
• You should get the following response:

{"Square Meters":11.4271,"Actual Median Income":54058,"Predicted Median Income":56388,"Percent Difference":4.31}

• Now that your local endpoint works, you can modify widgets/RIntegration/Widget.js to point at your own endpoint.

Look for the existing code:

$.ajax({
	type: 'POST',
	url: 'https://api2.retailscientifics.com/esri_demo/runmodel',
	data: {
		inputDataframe: JSON.stringify(dataframe),
		apikey: 'some_secret_key'
	},
...

and change it to

$.ajax({
	type: 'POST',
	url: 'http://localhost:8001/runmodel',
	data: {
		inputDataframe: JSON.stringify(dataframe),
		apikey: 'some_secret_key'
	},
...

• Run 'python3 server.py' to run the frontend component and visit https://localhost:8000 (bypassing any security warnings). You should now be able to submit the input form and receive output.

Deploying Remote Server

These instructions apply to linx-based servers - in particular, distributions such as Ubuntu or Debian. Instructions for other operating systems will differ slightly, but essentially just involve setting up a web server for the frontend and R for the backend on the remote server. Ensure you have root or sudo access on your server.

Running the R Model

This is the first step to productionalizing - you should be able to access the R server using your server's IP address and port.

• SSH into the remote server and install R (sudo apt-get install r-base)
• Start Server/esri_demo.R and attempt to navigate a browser to http://yourServerIP:8001/runmodel?inputDataframe={"Latitude":38.98730482188786,"Longitude":-114.84300214791408,"LocationName":"asdsa","LocationSquareFootage":"123","locationLatitudeLongitude":"38.98730,+-114.84300","PopulationDensity":"low","PropBoomers":"average","NeighborsToUse":"4"}&apikey=some_secret_key
  • You should receive the same output as when testing locally

Creating an Friendly API

For production applications, you generally want to supply a full domain name that can be accessed. The approach here will involve using an HTTP server such as nginx, which will listen for requests to a certain endpoint and forward them to your R model (usually referred to as 'reverse proxying').

• SSH into the remote server and install nginx (sudo apt get install nginx)
• Enable and start nginx using sudo systemctl enable nginx && sudo systemctl start nginx
• If there is a firewall installed on the server, ensure that ports 80 and 443 are open
  • A commonly used program is UFW, which allows you to simply run sudo ufw allow http && sudo ufw allow https && sudo ufw reload
  • You should now be able to visit your server's IP address in a browser and see the nginx test page.
• Modify the site configuration file for nginx using something like vim /etc/nginx/sites-enabled/default
• Use the file Server/nginx.config in this repository as a guide:
  • This forwards requests to http://yourserver/esri_demo to the port that R plumber listens on
  • The port number under upstream esridemo_backend should match the port used in Server/esri_demo.R
• Restart nginx with systemctl restart nginx
  • You should now be able to visit http://yourserver/esridemo/runmodel?inputDataframe={"Latitude":38.98730482188786,"Longitude":-114.84300214791408,"LocationName":"asdsa","LocationSquareFootage":"123","locationLatitudeLongitude":"38.98730,+-114.84300","PopulationDensity":"low","PropBoomers":"average","NeighborsToUse":"4"}&apikey=some_secret_key
  • You should again receive the same response as when testing locally