Have you ever wondered how Gmail autocompletes your sentences, or, what powers the WhatsApp suggestions when you’re typing a message? The technology behind these helpful writing hints is machine learning. In this course, you'll build and train machine learning models for different natural language generation tasks. For example, you'll train a model on the literary works of Shakespeare and generate text in the style of his writing. You'll also learn how to create a neural translation model to translate English sentences into French. Finally, you'll train a seq2seq model to generate your own natural language autocomplete sentences, just like Gmail!
The order of words in sentences is important (unless Yoda you are called). That’s why in this chapter, you’ll learn how to represent your data sequentially and use neural network architecture to model your text data. You'll learn how to create and train a recurrent network to generate new text, character by character. You'll also use the names dataset to build your own baby name generator, using a very simple recurrent neural network and the Keras package.
In this chapter, you’ll find out how to overcome the limitations of recurrent neural networks when input sequences span long intervals. To avoid vanishing and exploding gradient problems you'll be introduced to long short term memory (LSTM) networks that are more effective when working with long-term dependencies. You'll work on a fun project where you'll build and train a simple LSTM model using selected literary works of Shakespeare to generate new text in the unique writing style of Shakespeare.
In this chapter, you'll learn about the encoder-decoder architecture and how it can be used to model sequence-to-sequence datasets, converting information from one domain to another domain. You'll use this knowledge to build a model for neural machine translation, training your model to translate English sentences into French.
In this chapter, you'll build your very own machine learning seq2seq model. You'll use real-world messages from the Enron email dataset to train an encoder-decoder model. Using this you’ll predict the correct ending for an incomplete input sentence.
In the following tracksDeep Learning for NLP
“I've used other sites—Coursera, Udacity, things like that—but DataCamp's been the one that I've stuck with.”
Devon Edwards Joseph
Lloyds Banking Group
“DataCamp is the top resource I recommend for learning data science.”
Harvard Business School
“DataCamp is by far my favorite website to learn from.”
Decision Science Analytics, USAA