Overview

The tfhub package provides R wrappers to TensorFlow Hub.

TensorFlow Hub is a library for reusable machine learning modules.

TensorFlow Hub is a library for the publication, discovery, and consumption of reusable parts of machine learning models. A module is a self-contained piece of a TensorFlow graph, along with its weights and assets, that can be reused across different tasks in a process known as transfer learning. Transfer learning can:

  • Train a model with a smaller dataset,
  • Improve generalization, and
  • Speed up training.

Installation

You can install the released version of tfhub from CRAN with:

install.packages("tfhub")

And the development version from GitHub with:

# install.packages("devtools")
devtools::install_github("rstudio/tfhub")

After installing the tfhub package you need to install the TensorFlow Hub python module:

library(tfhub)
install_tfhub()

Loading modules

Modules can be loaded from URL’s and local paths using hub_load()

module <- hub_load("https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/2")

Module’s behave like functions and can be called with Tensors eg:

input <- tf$random$uniform(shape = shape(1,224,224,3), minval = 0, maxval = 1)
output <- module(input)

Using with Keras

The easiest way to get started with tfhub is using layer_hub. A Keras layer that loads a TensorFlow Hub module and prepares it for using with your model.

library(tfhub)
library(keras)

input <- layer_input(shape = c(32, 32, 3))

output <- input %>%
  # we are using a pre-trained MobileNet model!
  layer_hub(handle = "https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/2") %>%
  layer_dense(units = 10, activation = "softmax")

model <- keras_model(input, output)

model %>%
  compile(
    loss = "sparse_categorical_crossentropy",
    optimizer = "adam",
    metrics = "accuracy"
  )

We can then fit our model in the CIFAR10 dataset:

cifar <- dataset_cifar10()
cifar$train$x <- tf$image$resize(cifar$train$x/255, size = shape(224,224))

model %>%
  fit(
    x = cifar$train$x,
    y = cifar$train$y,
    validation_split = 0.2,
    batch_size = 128
  )

Using with tfdatasets

tfhub can also be used with tfdatasets:

  • hub_text_embedding_column()
  • hub_sparse_text_embedding_column()
  • hub_image_embedding_column()

Using with recipes

tfhub adds a step_pretrained_text_embedding that can be used with the recipes package.

An example can be found here.

tfhub.dev

tfhub.dev is a gallery of pre-trained model ready to be used with TensorFlow Hub.