xiangdonggu/flaskrpy

Simple API Server For R User

Simple API Server For R User

We provide a simple Python Flask server for R users, who can easily deploy models or functions written in R to the server.

How does it work?

A simple web server is started by Python Flask, which also starts an R process. The server translates any API call into R function and return the result in JSON format.

Install

Install R package

devtools::install_github("xiangdonggu/flaskrpy")

Flask web app

We can simply create a new directory and put app.py and .Rprofile files there. .Rprofile runs initial pieces of R codes such as loading necessary packages when the R process is started. It can be modified to suit user's needs.

Start the server

Go the directory that contains app.py and .Rprofile

python app.py

The server is up and running and we can now interact with it in R.

Example

Below we illustrate how to deploy models to R and make API calls. The API can be called by any platform that supports web API call without the needs of R.

library(flaskrpy)
library(jsonlite)

# Build a random forest model, with transformation function
library(randomForest)

# We do not need to bin the variable, we just want to test
# if we can deploy the whole working environment with
# a lot of objects dependencies
cuts <- c(-Inf, 5.1, 5.8, 6.4, Inf)

trans <- function(data) {
  data$Sepal.Length <- cut(data$Sepal.Length, cuts)
  data
}

fit <- randomForest(Species~., data = trans(iris))

# prediction function for API to use, with dependencies
# on cuts, transform, fit
pred <- function(d) {
  d <- as.data.frame(d, stringsAsfactors = FALSE)
  response <- predict(fit, newdata = trans(d))
  list(prediction = response)
}

# We need to make pred function exposed, user should explicitly
# declare which functions can be exposed to API
pred <- api_expose(pred)

# Deploy necessary objects to API server
api_deploy(pred, fit, cuts, trans, model_name = "iristest",
           host = "http://127.0.0.1:5000")

# Call API in R
api_call(model = "iristest", func = "pred", req = iris[1:5, ],
         host = "http://127.0.0.1:5000")

We can also call the deployed model prediction API using CURL

curl -i -k -H "Content-Type: application/json" -X POST -d 
'[{"Sepal.Length":5.1,"Sepal.Width":3.5,"Petal.Length":1.4,
"Petal.Width":0.2,"Species":"setosa"},{"Sepal.Length":4.9,
"Sepal.Width":3,"Petal.Length":1.4,"Petal.Width":0.2,
"Species":"setosa"},{"Sepal.Length":4.7,"Sepal.Width":3.2,
"Petal.Length":1.3,"Petal.Width":0.2,"Species":"setosa"},
{"Sepal.Length":4.6,"Sepal.Width":3.1,"Petal.Length":1.5,
"Petal.Width":0.2,"Species":"setosa"},{"Sepal.Length":5,
"Sepal.Width":3.6,"Petal.Length":1.4,"Petal.Width":0.2,
"Species":"setosa"}]' http://127.0.0.1:5000/r/iristest/pred