We’re happy to announce that the primary model of tfhub is now on CRAN. tfhub is an R interface to TensorFlow Hub – a library for the publication, discovery, and consumption of reusable elements of machine studying fashions. A module is a self-contained piece of a TensorFlow graph, together with its weights and belongings, that may be reused throughout completely different duties in a course of often known as switch studying.

The CRAN model of tfhub could be put in with:

After putting in the R bundle it is advisable set up the TensorFlow Hub python bundle. You are able to do it by working:

Getting began

The important operate of tfhub is layer_hub which works similar to a keras layer however lets you load a whole pre-trained deep studying mannequin.

For instance you’ll be able to:

library(tfhub)
layer_mobilenet <- layer_hub(
  deal with = "https://tfhub.dev/google/tf2-preview/mobilenet_v2/classification/4"
)

It will obtain the MobileNet mannequin pre-trained on the ImageNet dataset. tfhub fashions are cached regionally and don’t must be downloaded the following time you utilize the identical mannequin.

Now you can use layer_mobilenet as a traditional Keras layer. For instance you’ll be able to outline a mannequin:

library(keras)
enter <- layer_input(form = c(224, 224, 3))
output <- layer_mobilenet(enter)
mannequin <- keras_model(enter, output)
abstract(mannequin)

Mannequin: "mannequin"
____________________________________________________________________
Layer (kind)                  Output Form               Param #    
====================================================================
input_2 (InputLayer)          [(None, 224, 224, 3)]      0          
____________________________________________________________________
keras_layer_1 (KerasLayer)    (None, 1001)               3540265    
====================================================================
Complete params: 3,540,265
Trainable params: 0
Non-trainable params: 3,540,265
____________________________________________________________________

This mannequin can now be used to foretell Imagenet labels for a picture. For instance, let’s see the outcomes for the well-known Grace Hopper’s photograph:

img <- image_load("https://blogs.rstudio.com/tensorflow/posts/photographs/grace-hopper.jpg", target_size = c(224,224)) %>% 
  image_to_array()
img <- img/255
dim(img) <- c(1, dim(img))
pred <- predict(mannequin, img)
imagenet_decode_predictions(pred[,-1,drop=FALSE])[[1]]

  class_name class_description    rating
1  n03763968  military_uniform 9.760404
2  n02817516          bearskin 5.922512
3  n04350905              swimsuit 5.729345
4  n03787032       mortarboard 5.400651
5  n03929855       pickelhaube 5.008665

TensorFlow Hub additionally provides many different pre-trained picture, textual content and video fashions.
All attainable fashions could be discovered on the TensorFlow hub web site.

Yow will discover extra examples of layer_hub utilization within the following articles on the TensorFlow for R web site:

Utilization with Recipes and the Characteristic Spec API

tfhub additionally provides recipes steps to make
it simpler to make use of pre-trained deep studying fashions in your machine studying workflow.

For instance, you’ll be able to outline a recipe that makes use of a pre-trained textual content embedding mannequin with:

rec <- recipe(obscene ~ comment_text, knowledge = practice) %>%
  step_pretrained_text_embedding(
    comment_text,
    deal with = "https://tfhub.dev/google/tf2-preview/gnews-swivel-20dim-with-oov/1"
  ) %>%
  step_bin2factor(obscene)

You’ll be able to see a whole working instance right here.

You too can use tfhub with the brand new Characteristic Spec API carried out in tfdatasets. You’ll be able to see a whole instance right here.

We hope our readers have enjoyable experimenting with Hub fashions and/or can put them to good use. In case you run into any issues, tell us by creating a difficulty within the tfhub repository

Reuse

Textual content and figures are licensed underneath Inventive Commons Attribution CC BY 4.0. The figures which have been reused from different sources do not fall underneath this license and could be acknowledged by a notice of their caption: “Determine from …”.

Quotation

For attribution, please cite this work as

Falbel (2019, Dec. 18). Posit AI Weblog: tfhub: R interface to TensorFlow Hub. Retrieved from https://blogs.rstudio.com/tensorflow/posts/2019-12-18-tfhub-0.7.0/

BibTeX quotation

@misc{tfhub,
  creator = {Falbel, Daniel},
  title = {Posit AI Weblog: tfhub: R interface to TensorFlow Hub},
  url = {https://blogs.rstudio.com/tensorflow/posts/2019-12-18-tfhub-0.7.0/},
  12 months = {2019}
}