R Interface to AutoKeras

Travis Build Status Coverage status lifecycle

AutoKeras is an open source software library for automated machine learning (AutoML). It is developed by DATA Lab at Texas A&M University and community contributors. The ultimate goal of AutoML is to provide easily accessible deep learning tools to domain experts with limited data science or machine learning background. AutoKeras provides functions to automatically search for architecture and hyperparameters of deep learning models.

Check out the AutoKeras blogpost at the RStudio TensorFlow for R blog.



AutoKeras is currently only available as a GitHub package. To install it run the following from an R console:

if (!require("remotes")) {

Then, use the install_autokeras() function to install TensorFlow:



autokeras R package has a configured Docker image.

Steps to run it:

From a bash console:

docker pull jcrodriguez1989/r-autokeras:1.0.0
docker run -it jcrodriguez1989/r-autokeras:1.0.0 /bin/bash

To run the docker image, and share the current folder (in home machine) to the /data path (in the docker machine), then do:

docker run -it -v ${PWD}:/data jcrodriguez1989/r-autokeras:1.0.0 /bin/bash
ls /data # once when the docker image is running


CIFAR-10 dataset


# Get CIFAR-10 dataset, but not preprocessing needed
cifar10 <- dataset_cifar10()
c(x_train, y_train) %<-% cifar10$train
c(x_test, y_test) %<-% cifar10$test

# Create an image classifier, and train 10 different models
clf <- model_image_classifier(
  directory = ".",
  name = "autokeras_cifar10",
  max_trials = 10
) %>%
  fit(x_train, y_train)
# And use it to evaluate, predict
clf %>% evaluate(x_test, y_test)
clf %>% predict(x_test[1:10, , , ])
# Get the best trained Keras model, to work with the keras R library
# Actually, exporting the model as a Keras model is not working as expected,
# check out the bug https://github.com/keras-team/autokeras/issues/929 .
(keras_model <- export_model(clf))
# Save the trained AutoKeras model
save_model(clf, "~/my_model.pkl")
# Later, the AutoKeras model can be loaded again with:
my_model <- load_model("~/my_model.pkl")

IMDb dataset


# Get IMDb dataset
imdb <- dataset_imdb(num_words = 1000)
c(x_train, y_train) %<-% imdb$train
c(x_test, y_test) %<-% imdb$test

# AutoKeras procceses each text data point as a character vector,
# i.e., x_train[[1]] "<START> this film was just brilliant casting..",
# so we need to transform the dataset.
word_index <- dataset_imdb_word_index()
word_index <- c(
  "<PAD>", "<START>", "<UNK>", "<UNUSED>",
x_train <- lapply(x_train, function(x) {
  paste(word_index[x + 1], collapse = " ")
x_test <- lapply(x_test, function(x) {
  paste(word_index[x + 1], collapse = " ")

x_train <- array(unlist(x_train))
x_test <- array(unlist(x_test))
y_train <- matrix(y_train, ncol = 1)
y_test <- matrix(y_test, ncol = 1)

# Create a text classifier, and train 10 different models
clf <- model_text_classifier(max_trials = 10) %>%
  fit(x_train, y_train)
# And use it to evaluate, predict
clf %>% evaluate(x_test, y_test)
clf %>% predict(x_test[1:10])
# Get the best trained Keras model, to work with the keras R library