Deep Learning with PyTorch Quick Start Guide

Learn to train and deploy neural network models in Python

David Julian

• Editore: Packt Publishing
• Anno: 2018
• ISBN:
  9781789539738
• Edizione: 1
• Argomento: Computer Science
• Categoria: Neural Networks

Introduction to deep learning and PyTorch by building a convolutional neural network and recurrent neural network for real-world use cases such as image classification, transfer learning, and natural language processing.

Key Features

• Clear and concise explanations
• Gives important insights into deep learning models
• Practical demonstration of key concepts

Book Description

PyTorch is extremely powerful and yet easy to learn. It provides advanced features, such as supporting multiprocessor, distributed, and parallel computation. This book is an excellent entry point for those wanting to explore deep learning with PyTorch to harness its power.

This book will introduce you to the PyTorch deep learning library and teach you how to train deep learning models without any hassle. We will set up the deep learning environment using PyTorch, and then train and deploy different types of deep learning models, such as CNN, RNN, and autoencoders.

You will learn how to optimize models by tuning hyperparameters and how to use PyTorch in multiprocessor and distributed environments. We will discuss long short-term memory network (LSTMs) and build a language model to predict text.

By the end of this book, you will be familiar with PyTorch's capabilities and be able to utilize the library to train your neural networks with relative ease.

What you will learn

• Set up the deep learning environment using the PyTorch library
• Learn to build a deep learning model for image classification
• Use a convolutional neural network for transfer learning
• Understand to use PyTorch for natural language processing
• Use a recurrent neural network to classify text
• Understand how to optimize PyTorch in multiprocessor and distributed environments
• Train, optimize, and deploy your neural networks for maximum accuracy and performance
• Learn to deploy production-ready models

Who this book is for

Developers and Data Scientist familiar with Machine Learning but new to deep learning, or existing practitioners of deep learning who would like to use PyTorch to train their deep learning models will find this book to be useful. Having knowledge of Python programming will be an added advantage, while previous exposure to PyTorch is not needed.

Deep Learning Fundamentals

Deep learning is generally considered a subset of machine learning, involving the training of artificial neural networks (ANNs). ANNs are at the forefront of machine learning. They have the ability to solve complex problems involving massive amounts of data. Many of the principles of machine learning generally are also important in deep learning specifically, so we will spend some time reviewing these here.

In this chapter, we will discuss the following topics:

• Approaches to machine learning
• Learning tasks
• Features
• Models
• Artificial neural networks

Approaches to machine learning

Prior to general machine learning, if we wanted to, for example, build a spam filter, we could start by compiling a list of words that commonly appear in spam. The spam detector then scans each email and when the number of blacklisted words reaches a threshold, the email would be classified as spam. This is called a rules-based approach, and is illustrated in the following diagram:

The problem with this approach is that once the writers of spam know the rules, they are able to craft emails that avoid this filter. The people with the unenviable task of maintaining this spam filter would have to continually update the list of rules. With machine learning, we can effectively automate this rule-updating process. Instead of writing a list of rules, we build and train a model. As a spam detector, it will be more accurate since it can analyze large volumes of data. It is able to detect patterns in data that would be impossible for a human to do in a meaningful timeframe. The following diagram illustrates this approach:

There are a large number of ways that we can approach machine learning and these approaches are broadly characterized by the following factors:

• Whether or not models are trained with labelled training data. There are several possibilities here, including entirely supervised, semi-supervised, based on reinforcement, or entirely unsupervised.
• Whether they are online (that is, learning on the fly as new data is presented), or learn using pre-existing data. This is referred to as batch learning.
• Whether they are instance-based, simply comparing new data to known data, or model-based, involving the detection of patterns and building a predictive model.

These approaches are not mutually exclusive and most algorithms are a combination of the approaches. For example, a typical way to build a spam detector is using an online, model-based supervised learning algorithm.

Learning tasks

There are several distinct types of learning tasks that are partially defined by the type of data that they work on. Based on this, we can divide learning tasks into two broad categories:

• Unsupervised learning: Data is unlabeled so the algorithm must infer a relationship between variables or by finding clusters of similar variables
• Supervised learning: Uses a labeled dataset to build an inferred function that can be used to predict the label of an unlabeled sample

Whether the data is labeled or not has a predetermining effect on the way a learning algorithm is built.

Unsupervised learning

One of the main drawbacks to supervised learning is that it requires data that is accurately labeled. Most real-world data consists of unlabeled and unstructured data and this is the major challenge to machine learning and the broader endeavor of artificial intelligence. Unsupervised learning plays an important role in finding structure in unstructured data. The division between supervised and unsupervised learning is not absolute. Many unsupervised algorithms are used to together with supervised learning; for example, where data is only partially labeled or when we are trying to find the most important features of a deep learning model.

Clustering

This is the most straightforward unsupervised method. In many cases, it does not matter that the data is unlabeled; what we are interested in is the fact that the data clusters around certain points. Recommender systems that, say, recommend movies or books from an online store often use clustering techniques. An approach here is for an algorithm to analyze a customer's purchase history, comparing it to other customers, and making recommendations based on similarities. The algorithm clusters customers' usage patterns into groups. At no time does the algorithm know what the groups are; it is able to work this out for itself. One of the most used clustering algorithms is k-means. This algorithm works by establishing cluster centers bas...

Indice dei contenuti

• Title Page
• Copyright and Credits
• About Packt
• Contributors
• Preface
• Introduction to PyTorch
• Deep Learning Fundamentals
• Computational Graphs and Linear Models
• Convolutional Networks
• Other NN Architectures
• Getting the Most out of PyTorch
• Other Books You May Enjoy