Deep Studying with R, 2nd Version

As we speak we’re happy to announce the launch of Deep Studying with R,
2nd Version. In comparison with the primary version,
the guide is over a 3rd longer, with greater than 75% new content material. It’s
not a lot an up to date version as a complete new guide.

This guide exhibits you methods to get began with deep studying in R, even when
you haven’t any background in arithmetic or knowledge science. The guide covers:

• Deep studying from first ideas

• Picture classification and picture segmentation

• Time collection forecasting

• Textual content classification and machine translation

• Textual content era, neural model switch, and picture era

Solely modest R data is assumed; every thing else is defined from
the bottom up with examples that plainly show the mechanics.
Study gradients and backpropogation—by utilizing tf$GradientTape()
to rediscover Earth’s gravity acceleration fixed (9.8 (m/s^2)). Study
what a keras Layer is—by implementing one from scratch utilizing solely
base R. Study the distinction between batch normalization and layer
normalization, what layer_lstm() does, what occurs if you name
match(), and so forth—all via implementations in plain R code.

Each part within the guide has obtained main updates. The chapters on
pc imaginative and prescient achieve a full walk-through of methods to strategy a picture
segmentation process. Sections on picture classification have been up to date to
use {tfdatasets} and Keras preprocessing layers, demonstrating not simply
methods to compose an environment friendly and quick knowledge pipeline, but additionally methods to
adapt it when your dataset requires it.

The chapters on textual content fashions have been utterly reworked. Learn to
preprocess uncooked textual content for deep studying, first by implementing a textual content
vectorization layer utilizing solely base R, earlier than utilizing
keras::layer_text_vectorization() in 9 other ways. Study
embedding layers by implementing a customized
layer_positional_embedding(). Study concerning the transformer structure
by implementing a customized layer_transformer_encoder() and
layer_transformer_decoder(). And alongside the way in which put all of it collectively by
coaching textual content fashions—first, a movie-review sentiment classifier, then,
an English-to-Spanish translator, and at last, a movie-review textual content
generator.

Generative fashions have their very own devoted chapter, masking not solely
textual content era, but additionally variational auto encoders (VAE), generative
adversarial networks (GAN), and magnificence switch.

Alongside every step of the way in which, you’ll discover sprinkled intuitions distilled
from expertise and empirical commentary about what works, what
doesn’t, and why. Solutions to questions like: when must you use
bag-of-words as a substitute of a sequence structure? When is it higher to
use a pretrained mannequin as a substitute of coaching a mannequin from scratch? When
must you use GRU as a substitute of LSTM? When is it higher to make use of separable
convolution as a substitute of standard convolution? When coaching is unstable,
what troubleshooting steps must you take? What are you able to do to make
coaching quicker?

The guide shuns magic and hand-waving, and as a substitute pulls again the curtain
on each crucial basic idea wanted to use deep studying.
After working via the fabric within the guide, you’ll not solely know
methods to apply deep studying to widespread duties, but additionally have the context to
go and apply deep studying to new domains and new issues.

Deep Studying with R, Second Version

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 might be acknowledged by a notice of their caption: “Determine from …”.

Quotation

For attribution, please cite this work as

Kalinowski (2022, Could 31). Posit AI Weblog: Deep Studying with R, 2nd Version. Retrieved from https://blogs.rstudio.com/tensorflow/posts/2022-05-31-deep-learning-with-R-2e/

BibTeX quotation

@misc{kalinowskiDLwR2e,
  writer = {Kalinowski, Tomasz},
  title = {Posit AI Weblog: Deep Studying with R, 2nd Version},
  url = {https://blogs.rstudio.com/tensorflow/posts/2022-05-31-deep-learning-with-R-2e/},
  12 months = {2022}
}