Problem installing keras on RStudio Cloud

[Leon]

Running the following setup script, which has worked before, but alas no longer...

The installation seems succesful, but when trying to fetch the mnist dataset, I simply get the error:

> mnist <- dataset_mnist()
Error:

Setup:

# Install packages --------------------------------------------------------
install.packages("tidyverse")
install.packages("devtools")


# Load libraries ----------------------------------------------------------
library("tidyverse")
library("devtools")


# Install keras / tensorflow ----------------------------------------------
install_github("rstudio/keras")
# Would you like to install miniconda? Y
library(keras)
install_keras(tensorflow = "1.13.1")


# Test installation -------------------------------------------------------
library("keras")
mnist <- dataset_mnist()
print(head(mnist))


# Clean up ----------------------------------------------------------------
rm(mnist)

[Leon]

Googling back and forth, I have now arrived at this setup, which works.

install.packages("remotes")
remotes::install_github("rstudio/tensorflow")
install.packages("reticulate")
library(tensorflow)
install_tensorflow(version = "2.0.0b1", method = "virtualenv", envname = "r-reticulate")
install.packages("keras")
library(keras)

The difference is not entirely clear to me though...

[javierluraschi]

We don't recommend using install_keras(tensorflow = "1.13.1") anymore. Instead, in rstudio.cloud use:

install.packages("keras")
keras::install_keras(tensorflow = "cpu")

However, please notice that tools like TensorFlow and Keras are not likely to run in the free version since it comes with 1GB of memory which is not sufficient to start these packages at the moment.