Learning Path: TensorFlow: Machine & Deep Learning Solutions

Harness the power of machine and deep learning of TensorFlow with ease

Created byPackt Publishing

Last updated 11/2017

English

What you'll learn

Deep diving into training, validating, and monitoring training performance
Set up and run cross-sectional examples (images, time-series, text, audio)
Load, interact, dissect, process, and save complex datasets
Predict the outcome of a simple time series using linear regression modeling
Resolve character-recognition problems using the recurrent neural network model
Work with Docker and Keras

Course content

3 sections • 85 lectures • 5h 23m total length

The Course Overview3:48
This video will provide an overview of the entire course.
Introducing Deep Learning3:59
Get the user excited about ML; a quick view of various platforms and key products using the ML tools (photo search, Siri, and so on).
Introduce TensorFlow
Installing TensorFlow on Mac OS X3:50
We will install the TensorFlow platform and associated prerequisites on the Mac OS X operating system.

Install the Python package manager PIP

Install dependencies such as SIX

Install TensorFlow
Installation on Windows – Pre-Reqeusite Virtual Machine Setup2:49
Since TensorFlow does not function natively on Windows, we will cover Virtual Machine setup so Windows users can run Linux on a VM.

Set up a virtual machine

Allocate the right amount of space for our installation and projects

Obtain and install an appropriate Linux installation image
Installation on Windows/Linux4:00
We will cover TensorFlow installation on Linux. This also covers installation on Windows via a Virtual Machine running Linux.

Set up a virtual machine

Obtain an appropriate Linux installation image
The Hand-Written Letters Dataset3:01
Which dataset do we use for our first classifier, and what attributes do we consider? Let's use a dataset of letters in numerous typefaces.

Find a suitable dataset for training our classifier

Review dataset attributes

Compare between MNIST and notMNIST datasets
Automating Data Preparation3:20
We just spent a good amount of time obtaining and preparing data, are we expected to do this for each run? No, you'll now learn to automate the initial steps used to feed the machine learning process.

Download data automatically

Extract data automatically

Load training and test images into a processable format
Understanding Matrix Conversions5:34
How do we represent images internally within the machine learning environment? We'll hold image data in a stack of matrices.

Load images into memory

Shuffle the images

Segment out distinct working sets from our images
The Machine Learning Life Cycle1:51
So now that we have looked upon the major concepts, the question remains how do we actually train the classifier in code? We are now going to code and train a classifier using TensorFlow.

Understand the life cycle

Design the neural network

Train the system
Reviewing Outputs and Results2:51
Once training is complete, how are the results interpreted and measured? We'll use the training set, with known labels to measure how well our trained system performs.

Use a trained network to predict classes for the test set

Use validation methods to compare predictions against actuals

Find the accuracy of our classifier on the test set
Getting Started with TensorBoard5:08
How do we monitor the internals of our training setup and execution? We'll use TensorBoard to view our network architecture and probe values through the training.

Get TensorBoard up and running

Explorer Events and Histograms functionality

Learn Graph Explorer
TensorBoard Events and Histograms5:21
How do we monitor the scalars and tensors on TensorBoard? We'll set up code for log specific variables and TensorBoard reads these structured logs.

Use specific commands to capture interesting variables in a structured manner

Point TensorBoard toward these variables

Browse through, view, and work with individual variable graphs
The Graph Explorer5:08
How do we view the network architecture we've built for our training? We'll push the graph to a TensorBoard readable log and view it on Graph Explorer.

Use specific code, push the graph to TensorBoard logs

View the network architecture on Graph Explorer

Zoom into individual components, view parameters, inputs, and outputs
Our Previous Project on TensorBoard5:02
How do we actually apply what you've just learned to our own project? You'll learn to go through individual changes to our previous project.

Initialize TensorBoard variables

Capture the graph as well as interesting variables

Execute training and launch TensorBoard to view outputs
Fully Connected Neural Networks4:44
How do we process images with more complex and better suited neural networks? For this purpose, we will use the well-known
convolutional neural networks also known as CNNs.

Learn the difference between fully connected networks and CNNs

Learn how CNNs work better than normal NNs

Change our previous code to add CNN model for training
Convolutional Neural Networks4:58
How do CNNs work and what does their model look like? We are going to talk about the basic model architecture of CNN.

Talk about what convolution is

Discuss what kernels are

Discuss CNNs architecture
Programming a CNN5:01
How to implement a basic CNN? We will look into the coding of a CNN.

Implement the model

Add appropriate layers

Talk about the code architecture in TensorFlow
Using TensorBoard on Our CNN1:58
How does it look after coding? We will visualize the implemented CNN in TensorFlow.

Add TensorBoard code at appropriate places

Run the graph with TensorBoard

Visualize it after the run
CNN Versus Fully Connected Network Performance2:08
How much better CNNs are as compared to normal fully connected networks? We will compare the performances of the two models on CIFAR-10 dataset.

Find test accuracy using FCN

Find accuracy using CNN

Compare the performance differences
Machine learning with TensorFlow

The Course Overview3:24
This video provides an overview of the entire course.
TensorFlow's Main Data Structure – Tensors7:14
Learn to know the basics of data management on tensors.

Explore and transform datac
Know the tensor data types
Interact with TensorFlow
Handling the Computing Workflow – TensorFlow's Data Flow Graph5:25
Learn how to handle the computing workflow of TensorFlow’s data flow graph.

Computation Graph Building

Variable initialization

Example Graph Building
Basic Tensor Methods8:22
Explore some basic methods supported by TensorFlow.

Tensor segmentation

Tensor shape transformations

Data flow structure and result visualization
How TensorBoard Works?5:32
Learn to know how TensorBoard works.

TensorFlow mechanism

Createthe TensorFlow graph

Interact with the TensorBoard's GUI
Reading Information from Disk4:00
Ability to read information from disk.

Read image data

Load and process the images
Learning from Data –Unsupervised Learning2:15
Learn to review two cases of unsupervised learning.

Represent the clustering algorithm

Simplify the data elements
Mechanics of k-Means3:34
Learn to know the mechanics of k-means.

Simplify the k-means graph

Algorithm iteration criterion
k-Nearest Neighbor5:33
Ability to know the k-nearest neighbors.

Nearest neighbor algorithm

How to round

How to bin numeric vectors
Project 1 – k-Means Clustering on Synthetic Datasets4:07
Learn the few topics about the clustering on synthetic datasets.

Generate the dataset

Results description
Project 2 – Nearest Neighbor on Synthetic Datasets1:52
Ability to load a dataset with which the k-means algorithm has problems separating classes.

Model architecture

Loss function description
Univariate Linear Modelling Function4:53
Learn to interact with linear equation using univariate linear modelling function.

Generatethe sample data

Minimize the cost function
Optimizer Methods in TensorFlow – The Train Module3:11
Learn to train the optimization stage which is a vital part of the machine learning workflow.

Optimize the gradients

Optimize the instance types
Univariate Linear Regression5:10
Ability to create a regression model that tries to fit a linear function that minimizes the error function.

Review the results with TensorBoard
Multivariate Linear Regression5:14
Ability to work on a regression problem involving more than one variable.

Optimize the loop – cost function
Logistic Function Predecessor – The Logit Functions4:07
Learn to review the original function on which it is based, and which gives it some of its more general properties.

Map to a probability distribution

Calculate the inverse of the logit function
The Logistic Function5:53
Learn to know the logistic function which will serve us to represent the binary options in our new regression tasks

Calculate the regression equation

Mappings of the uniclass and multiclass logistic regression

Smooth the error surface
Univariate Logistic Regression6:55
Ability to work approximating the probability of the presence of heart disease, using an univariate logistic regression.

Encode data structure

Create the initial variables

Create the linear model variables
Univariate Logistic Regression with skflow2:27
Explore the univariate examples domain which will use help from a new library, which eases the model building for us, called keras.

Classify softmax
Preliminary Concepts7:41
Learn to build a simple framework into the neural network components and architectures.

Activatethe neural network frameworks

Minimize the loss function

Preprocess sklearn
First Project – Non-Linear Synthetic Function Regression2:31
Learn to know how close we can be predicting values taken from a test population

Dataset preprocessing

Dataset description and loading
Second Project – Modeling Cars Fuel Efficiency with Non-Linear Regression3:05
Learn to build a regression model for the fuel efficiency of several car models, based on several variables

Decrease the global variable
Third Project – Learning to Classify Wines: Multiclass Classification2:56
Ability to work with a more complex dataset

Optimize the loss function

Optimizeloss description
Origin of Convolutional Neural Networks3:26
Learn to know the historical development of the operation and then start looking at convolution in the continuous domain.

Apply the concept of convolution in the discrete domain

Interpret the convolution operations
Applying Convolution in TensorFlow3:55
Ability to apply convolution in TensorFlow.

Apply convolution to a grayscale image
Subsampling Operation –Pooling2:56
Ability to get the maximum and the average of the elements for an applied kernel.

Apply pool operations in TensorFlow
Improving Efficiency – Dropout Operation2:15
Ability to know how the dropout operation reduces the value of some randomly selected weights to zero.

Apply the dropout operation in TensorFlow
Convolutional Type Layer Building Methods1:02
Learn to use simple utility functions to facilitate the building of convolutional layers.

Maintain the initial parameters of the layer
MNIST Digit Classification3:30
Learn to work for the first time on one of the most well-known datasets for pattern recognition.

Dataset description and loading

Dataset preprocessing
Image Classification with the CIFAR10 Dataset2:27
Learn to work on one of the most extensively used datasets in image comprehension.

Produce the scoring of the trained model
Recurrent Neural Networks3:39
Define a sequential model of neural networks, which have the property of reusing information already given

Explode and vanish the gradients

LSTM neural networks
A Fundamental Component – Gate Operation and Its Steps4:23
Ability to understand the building blocks of the internal of the LSTM cell and also we will describe the main operational block of the LSTM.

Set up values to the input gate

Set up values to change state

Filter the output cell state
TensorFlow LSTM Useful Classes and Methods2:01
Learn to review the main classes and methods that we can use to build a LSTM layer.

Reduce the loss of information for the initial training

Use a stack of connected cells
Univariate Time Series Prediction with Energy Consumption Data2:36
Gain the ability to solve a problem of the domain of regression.

Dataset preprocessing – consumption data

Modeling architecture
Writing Music "a la" Bach8:06
Learn to work with a recurrent neural network specialized in character sequences, or the char RNN model.

Network training

Character level models
Deep Neural Network Definition and Architectures Through Time2:34
Exploring to the Neural Network architectures with like tens of layer, or combinations of complex constructs.

Review the milestone architectures

Experience a blast on news techniques
Alexnet3:52
Learn to extend the complexity of the models.

Batch normalization
Inception V30:59
Learn to illustrate how the improved inception module can be interpreted.

Reduce the number of convolutions

Increase the general depth of the networks
Residual Networks (ResNet)2:05
Achieve the ability to use the output of each constitutional layer, and also combine the output of the layer with the original input.

Reduce the parameter number and complexity of Alexnet

Organize the structural layout as a truncated fractal
Painting with Style – VGG Style Transfer3:10
Learn to work with the implementation of the paper A Neural Algorithm of Artistic Style from Leon Gatys.

Load the parameters files with scipy.io.loadmat

Dataset preprocessing
Windows Installation2:38
Gain the ability to develop TensorFlow with the Windows operating system.
Download the Docker toolbox installer
Create the Docker machine
MacOS Installation2:57
Learn to install TensorFlow on mac OS .
Install the TensorFlow package
Import the TensorFlow module
Building Machine Learning Systems with TensorFlow

The Course Overview2:52
This video provides an overview of the entire course.
Installing Docker4:09
How to download and install Docker on Linux and Windows?

Learn how to download the Docker installer

Explore the installation process of Docker on Windows

Learn the script Docker installation process on Linux
The Machine Learning Dockerfile4:09
How to write a Dockerfile that configures the packages to get Keras running in a Docker container?

Clone the video source code

Code the base image and needed Linux packages

Expose ports and volumes to allow you container to talk to the world
Sharing Data2:59
How to configure Docker security settings on Windows to allow access to Jupyter notebooks?

Configure needed security settings in Docker

Test security to validate volume sharing

Run the machine learning Dockerfile
Machine Learning REST Service3:15
How to expose Jupyter as a machine learning REST service from your Docker container?

Run the Docker container

Use the exposed service endpoint URL

Verify Keras installation and loading
MNIST Digits3:16
How to encode the images for machine learning?

Download the MNIST digits

Treat images as numbers

Visualize the encoded images
Tensors: Just Multidimensional Arrays2:19
In this video, we will understand what a tensor is and we will Learn how to create them.

Use NumPy multidimensional arrays to store tensors

Set the tensor shapes

Set the tensor datatypes
Turning Images into Tensors2:20
This video explains how to encode images into tensors, a needed step for machine learning.

Encode images as floating point

Understand samples and their placement in tensors

Normalize data to facilitate learning algorithms
Turning Categories into Tensors2:48
How to encode categories into tensors?

Understand classes and discreet predictions for categories

Use one-hot encoding to create tensors

Visualize one hot encoding as a bitmap
Classical/Dense Neural Network2:46
What is the structure of a neural network?

Use code to draw a dense neural network

Use code to draw a neural network designed for MNIST digits

Understand the relationship between tensors and networks
Activation and Non Linearity2:25
What is non-linearity and why does it matter for neural networks?

Use nonlinearity in networks

Use the sigmoid in networks

Use the ReLU in networks
Softmax2:59
What is softmax, and where do we use it in networks?

Understand the softmax function

Use softmax to predict classes or categories

Use softmax to predict probabilities
Training and Testing Data2:55
What are training and testing data sets?

Loading data into normalized data sets

Use one-hot encoding to create training and testing outputs

Understand the use of training versus testing data sets
Dropout and Flatten3:07
What are dropout and flatten layers and when do we use them in networks?

Use Keras functional model and input layers

Use the dropout layer to improve learning

Use the flatten layer to make layer and tensor shapes compatible
Solvers3:45
What are solvers and what is their role in networks?

Understand learning and optimization

Compile and fitting a model to data with a solver

Understand loss and accuracy to evaluate a solver’s fitting of data
Hyperparameters2:17
In this video, we will learn What are hyperparameters and parameters and we will also see a difference between them.

Understand trainable parameters in networks

Understand non- trainable parameters in networks

Understand hyperparameters in network layers
Grid Search6:30
What is grid search and when do we use it in training networks?

Understand the difference between optimization with a solver and grid search

Configure a hyperparameter grid

Perform grid search and report outcomes
Convolutions3:23
What are convolutions and why would we use them in networks?

Use code to draw convolutions

Understand convolutions in two dimensions

Understand the benefits of convolutions for performance
Pooling3:44
What is pooling, and why do we use it work convolutions?

Understand one dimensional pooling

Understand two dimensional pooling

Understand pooling across image channels
Convolutional Neural Network3:23
How do we create a convolutional network to learn to recognize images?

Understand MNIST digits and image channels

Construct a convolutional network

Train and predict results with a convolutional network
Deep Neural Network5:54
What is a deep neural network compared to a classical or convolutional network?

Understand the role of GPU in deep network performance

Build a multiple block, deep network

Train and predict results with a deep network
REST API Definition5:40
How do we define an API for use with machine learning?

Clone the source code for the REST service

Configure a declarative API with Swagger/ OpenAPI yaml

Code an API handler that can receive an image file and classify it with keras
Trained Models in Docker Containers4:47
How do we create a deployable REST service with a trained Keras model?

Understand the need for a trained model file along with the API service code

Code a Dockerfile that will build a Python runtime along with a pre-trained Keras model

Build the Docker container from the command line
Making Predictions3:17
How do we use our API to make digit predictions?

Run the Docker container mapping ports to create a service

Use the Swagger/ Connexion built in UI to test the API

Use the API from the command line with curl
Tensorflow Deep Learning Solutions for Images

Requirements

This Learning Path takes a step-by-step approach, helping you explore all the functioning of TensorFlow.

Description

Google's brainchild TensorFlow, in its first year, has more than 6000 open source repositories online. TensorFlow, an open source software library, is extensively used for numerical computation using data flow graphs.The flexible architecture allows you to deploy computation to one or more CPUs or GPUs in a desktop, server, or mobile device with a single API. So if you’re looking forward to acquiring knowledge on machine learning and deep learning with this powerful TensorFlow library, then go for this Learning Path.

Packt’s Video Learning Paths are a series of individual video products put together in a logical and stepwise manner such that each video builds on the skills learned in the video before it.

The highlights of this Learning Path are:

Setting up TensorFlow for actual industrial use, including high-performance setup aspects like multi-GPU support
Embedded with solid projects and examples to teach you how to implement TensorFlow in production
Empower you to go from concept to a production-ready machine learning setup/pipeline capable of real-world usage

Let's take a look at your learning journey. You will start by exploring unique features of the library such as data flow graphs, training, visualization of performance with TensorBoard – all within an example-rich context using problems from multiple industries. The focus is towards introducing new concepts through problems which are coded and solved over the course of each video. You will then learn how to implement TensorFlow in production. Each project in this Learning Path provides exciting and insightful exercises that will teach you how to use TensorFlow and show you how layers of data can be explored by working with tensors. Finally, you will be acquainted with the different paradigms of performing deep learning such as deep neural nets, convolutional neural networks, recurrent neural networks, and more, and how they can be implemented using TensorFlow.

On completion of this Learning Path, you will have gone through the full lifecycle of a TensorFlow solution with a practical demonstration to system setup, training, validation, to creating pipelines for real world data -- all the way to deploying solutions into a production settings.

Meet Your Expert:

We have the best works of the following esteemed authors to ensure that your learning journey is smooth:

Shams Ul Azeem is an undergraduate of NUST Islamabad, Pakistan in Electrical Engineering. He has a great interest in computer science field and started his journey from android development. Now he’s pursuing his career in machine learning, particularly in deep learning by doing medical related freelance projects with different companies. He was also a member of RISE lab, NUST and has a publication in IEEE International Conference, ROBIO as a co-author on “Designing of motions for humanoid goal keeper robots”.
Rodolfo Bonnin a systems engineer and PhD student at Universidad Tecnológica Nacional, Argentina. He also pursued Parallel Programming and Image Understanding postgraduate courses at Uni Stuttgart, Germany. He has done research on high-performance computing since 2005 and began studying and implementing convolutional neural networks in 2008, writing a CPU and GPU supporting the neural network feedforward stage. More recently he's been working in the field of fraud pattern detection with neural networks, and is currently working on signal classification using ML techniques.

Will Ballard serves as chief technology officer at GLG and is responsible for the Engineering and IT organizations. Prior to joining GLG, Will was the executive vice president of technology and engineering at Demand Media. He graduated Magna Cum Laude with a BS in Mathematics from Claremont McKenna College.

Who this course is for:

This Learning Path is aimed at data analysts, data scientists, and researchers who want to increase the speed and efficiency of their machine learning activities and results using TensorFlow.

Learning Path: TensorFlow: Machine & Deep Learning Solutions

What you'll learn

Explore related topics

Course content

Machine Learning with TensorFlow19 lectures • 1hr 15min

Building Machine Learning Systems with TensorFlow42 lectures • 2hr 44min

Tensorflow Deep Learning Solutions for Images24 lectures • 1hr 25min

Requirements

Description

Who this course is for: