Machine Learning for Developers电子书

Title Page

Copyright

Machine Learning for Developers

Credits

Foreword

About the Author

About the Reviewers

www.PacktPub.com

Why subscribe?

Customer Feedback

Preface

What this book covers

What you need for this book

Who this book is for

Conventions

Reader feedback

Customer support

Downloading the example code

Errata

Piracy

Questions

Introduction - Machine Learning and Statistical Science

Machine learning in the bigger picture

Types of machine learning

Grades of supervision

Supervised learning strategies - regression versus classification

Unsupervised problem solving–clustering

Tools of the trade–programming language and libraries

The Python language

The NumPy library

The matplotlib library

What's matplotlib?

Pandas

SciPy

Jupyter notebook

Basic mathematical concepts

Statistics - the basic pillar of modeling uncertainty

Descriptive statistics - main operations

Mean

Variance

Standard deviation

Probability and random variables

Events

Probability

Random variables and distributions

Useful probability distributions

Bernoulli distributions

Uniform distribution

Normal distribution

Logistic distribution

Statistical measures for probability functions

Skewness

Kurtosis

Differential calculus elements

Preliminary knowledge

In search of changes–derivatives

Sliding on the slope

Chain rule

Partial derivatives

Summary

The Learning Process

Understanding the problem

Dataset definition and retrieval

The ETL process

Loading datasets and doing exploratory analysis with SciPy and pandas

Working interactively with IPython

Working on 2D data

Feature engineering

Imputation of missing data

One hot encoding

Dataset preprocessing

Normalization and feature scaling

Normalization or standardization

Model definition

Asking ourselves the right questions

Loss function definition

Model fitting and evaluation

Dataset partitioning

Common training terms – iteration, batch, and epoch

Types of training – online and batch processing

Parameter initialization

Model implementation and results interpretation

Regression metrics

Mean absolute error

Median absolute error

Mean squared error

Classification metrics

Accuracy

Precision score, recall, and F-measure

Confusion matrix

Clustering quality measurements

Silhouette coefficient

Homogeneity, completeness, and V-measure

Summary

References

Clustering

Grouping as a human activity

Automating the clustering process

Finding a common center - K-means

Pros and cons of K-means

K-means algorithm breakdown

K-means implementations

Nearest neighbors

Mechanics of K-NN

Pros and cons of K-NN

K-NN sample implementation

Going beyond the basics

The Elbow method

Summary

References

Linear and Logistic Regression

Regression analysis

Applications of regression

Quantitative versus qualitative variables

Linear regression

Determination of the cost function

The many ways of minimizing errors

Analytical approach

Pros and cons of the analytical approach

Covariance/correlation method

Covariance

Correlation

Searching for the slope and intercept with covariance and correlation

Gradient descent

Some intuitive background

The gradient descent loop

Formalizing our concepts

Expressing recursion as a process

Going practical – new tools for new methods

Useful diagrams for variable explorations – pairplot

Correlation plot

Data exploration and linear regression in practice

The Iris dataset

Getting an intuitive idea with Seaborn pairplot

Creating the prediction function

Defining the error function

Correlation fit

Polynomial regression and an introduction to underfitting and overfitting

Linear regression with gradient descent in practice

Logistic regression

Problem domain of linear regression and logistic regression

Logistic function predecessor – the logit functions

Link function

Logit function

Logit function properties

The importance of the logit inverse

The sigmoid or logistic function

Properties of the logistic function

Multiclass application – softmax regression

Practical example – cardiac disease modeling with logistic regression

The CHDAGE dataset

Dataset format

Summary

References

Neural Networks

History of neural models

The perceptron model

Improving our predictions – the ADALINE algorithm

Similarities and differences between a perceptron and ADALINE

Limitations of early models

Single and multilayer perceptrons

MLP origins

The feedforward mechanism

The chosen optimization algorithm – backpropagation

Types of problem to be tackled

Implementing a simple function with a single-layer perceptron

Defining and graphing transfer function types

Representing and understanding the transfer functions

Sigmoid or logistic function

Playing with the sigmoid

Rectified linear unit or ReLU

Linear transfer function

Defining loss functions for neural networks

L1 versus L2 properties

Summary

References

Convolutional Neural Networks

Origin of convolutional neural networks

Getting started with convolution

Continuous convolution

Discrete convolution

Kernels and convolutions

Stride and padding

Implementing the 2D discrete convolution operation in an example

Subsampling operation (pooling)

Improving efficiency with the dropout operation

Advantages of the dropout layers

Deep neural networks

Deep convolutional network architectures through time

Lenet 5

Alexnet

The VGG model

GoogLenet and the Inception model

Batch-normalized inception V2 and V3

Residual Networks (ResNet)

Types of problem solved by deep layers of CNNs

Classification

Detection

Segmentation

Deploying a deep neural network with Keras

Exploring a convolutional model with Quiver

Exploring a convolutional network with Quiver

Implementing transfer learning

References

Summary

Recurrent Neural Networks

Solving problems with order — RNNs

RNN definition

Types of sequence to be modeled

Development of RNN

Training method — backpropagation through time

Main problems of the traditional RNNs — exploding and vanishing gradients

LSTM

The gate and multiplier operation

Part 1 — set values to forget (input gate)

Part 2 — set values to keep

Part 3 — apply changes to cell

Part 4 — output filtered cell state

Univariate time series prediction with energy consumption data

Dataset description and loading

Dataset preprocessing

Summary

References

Recent Models and Developments

GANs

Types of GAN applications

Discriminative and generative models

Reinforcement learning

Markov decision process

Decision elements

Optimizing the Markov process

Basic RL techniques: Q-learning

References

Summary

Software Installation and Configuration

Linux installation

Initial distribution requirements

Installing Anaconda on Linux

pip Linux installation method

Installing the Python 3 interpreter

Installing pip

Installing necessary libraries

macOS X environment installation

Anaconda installation

Installing pip

Installing remaining libraries via pip

Windows installation

Anaconda Windows installation

Summary