Probabilistic Deep Learning with Python 1st

Edition Oliver Duerr download

https://textbookfull.com/product/probabilistic-deep-learning-
with-python-1st-edition-oliver-duerr/

Download more ebook from https://textbookfull.com

We believe these products will be a great fit for you. Click
the link to download now, or visit textbookfull.com
to discover even more!

Deep Learning with Python 1st Edition François Chollet

https://textbookfull.com/product/deep-learning-with-python-1st-
edition-francois-chollet/

Deep Learning with Python MEAP 2nd Edition Francois

Chollet

https://textbookfull.com/product/deep-learning-with-python-
meap-2nd-edition-francois-chollet/

Deep Learning with Python Learn Best Practices of Deep

Learning Models with PyTorch 2nd Edition Nikhil Ketkar
Jojo Moolayil

https://textbookfull.com/product/deep-learning-with-python-learn-
best-practices-of-deep-learning-models-with-pytorch-2nd-edition-
nikhil-ketkar-jojo-moolayil/

Deep Learning with Python Learn Best Practices of Deep

Learning Models with PyTorch 2nd Edition Nikhil Ketkar
Jojo Moolayil

https://textbookfull.com/product/deep-learning-with-python-learn-
best-practices-of-deep-learning-models-with-pytorch-2nd-edition-
nikhil-ketkar-jojo-moolayil-2/
Deep Learning with Python Develop Deep Learning Models
on Theano and TensorFLow Using Keras Jason Brownlee

https://textbookfull.com/product/deep-learning-with-python-
develop-deep-learning-models-on-theano-and-tensorflow-using-
keras-jason-brownlee/

Advanced Data Analytics Using Python: With Machine

Learning, Deep Learning and NLP Examples Mukhopadhyay

https://textbookfull.com/product/advanced-data-analytics-using-
python-with-machine-learning-deep-learning-and-nlp-examples-
mukhopadhyay/

Applied Neural Networks with TensorFlow 2: API Oriented

Deep Learning with Python Orhan Gazi Yalç■n

https://textbookfull.com/product/applied-neural-networks-with-
tensorflow-2-api-oriented-deep-learning-with-python-orhan-gazi-
yalcin/

Computer Vision Using Deep Learning Neural Network

Architectures with Python and Keras 1st Edition Vaibhav
Verdhan

https://textbookfull.com/product/computer-vision-using-deep-
learning-neural-network-architectures-with-python-and-keras-1st-
edition-vaibhav-verdhan/

Learn Keras for Deep Neural Networks: A Fast-Track

Approach to Modern Deep Learning with Python 1st
Edition Jojo John Moolayil

https://textbookfull.com/product/learn-keras-for-deep-neural-
networks-a-fast-track-approach-to-modern-deep-learning-with-
python-1st-edition-jojo-john-moolayil/
With Python, Keras,
and TensorFlow Probability

Oliver Dürr
Beate Sick
with Elvis Murina

MANNING
 RNN1
CNN2 ZIP GAUS1 GAUS2
CNN1
FCNN1

Network Shelf Prob. Distribution Shelf

DATA

Like
Max

FCNN2 POIS
M

xL
a

ik
e

Data modeling with probabilistic DL. The network determines the parameters of a
probability distribution. Fit the model using the MaxLike principle. In the example shown,
the outcome is count data. Here it’s modeled by a Poisson distribution, where NN is used
to control its rate parameter λ (see the chosen last plate with one output node).
Probabilistic Deep Learning
 Probabilistic
Deep Learning
WITH PYTHON, KERAS, AND
TENSORFLOW PROBABILITY

OLIVER DÜRR
BEATE SICK
WITH ELVIS MURINA

MANNING
SHELTER ISLAND
For online information and ordering of this and other Manning books, please visit
www.manning.com. The publisher offers discounts on this book when ordered in quantity.
For more information, please contact
Special Sales Department
Manning Publications Co.
20 Baldwin Road
PO Box 761
Shelter Island, NY 11964
Email: orders@manning.com

©2020 by Manning Publications Co. All rights reserved.

No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in

any form or by means electronic, mechanical, photocopying, or otherwise, without prior written
permission of the publisher.

Many of the designations used by manufacturers and sellers to distinguish their products are
claimed as trademarks. Where those designations appear in the book, and Manning Publications
was aware of a trademark claim, the designations have been printed in initial caps or all caps.

Recognizing the importance of preserving what has been written, it is Manning’s policy to have
the books we publish printed on acid-free paper, and we exert our best efforts to that end.
Recognizing also our responsibility to conserve the resources of our planet, Manning books
are printed on paper that is at least 15 percent recycled and processed without the use of
elemental chlorine.

Development editor: Marina Michaels

Technical development editors: Michiel Trimpe and
Arthur Zubarev
Manning Publications Co. Review editor: Aleksandar Dragosavljević
20 Baldwin Road Production editor: Deirdre S. Hiam
PO Box 761 Copy editor: Frances Buran
Shelter Island, NY 11964 Proofreader: Keri Hales
Technical proofreader: Al Krinker
Typesetter: Dennis Dalinnik
Cover designer: Marija Tudor

ISBN: 9781617296079
Printed in the United States of America
 brief contents
PART 1 BASICS OF DEEP LEARNING .............................................1
1 ■ Introduction to probabilistic deep learning 3
2 ■ Neural network architectures 25
3 ■ Principles of curve fitting 62

PART 2 MAXIMUM LIKELIHOOD APPROACHES FOR

PROBABILISTIC DL MODELS .........................................91
4 ■ Building loss functions with the likelihood approach 93
5 ■ Probabilistic deep learning models with TensorFlow
Probability 128
6 ■ Probabilistic deep learning models in the wild 157

PART 3 BAYESIAN APPROACHES FOR PROBABILISTIC

DL MODELS. ..............................................................195
7 ■ Bayesian learning 197
8 ■ Bayesian neural networks 229

v
 contents
preface xi
acknowledgments xii
about this book xiv
about the authors xvii
about the cover illustration xviii

PART 1 BASICS OF DEEP LEARNING ...................................1

1 Introduction to probabilistic deep learning

1.1 A first look at probabilistic models 4
3

1.2 A first brief look at deep learning (DL) 6

A success story 8
1.3 Classification 8
Traditional approach to image classification 9 Deep learning
■

approach to image classification 12 Non-probabilistic

■

classification 14 Probabilistic classification 14

■

Bayesian probabilistic classification 16

1.4 Curve fitting 16
Non-probabilistic curve fitting 17 Probabilistic curve
■

fitting 18 Bayesian probabilistic curve fitting 20

■

vii
viii CONTENTS

1.5 When to use and when not to use DL? 21

When not to use DL 21 When to use DL 22
■ ■
When to use
and when not to use probabilistic models? 22
1.6 What you’ll learn in this book 23

2 Neural network architectures

2.1
25
Fully connected neural networks (fcNNs) 26
The biology that inspired the design of artificial NNs 26
Getting started with implementing an NN 28 Using a fully ■

connected NN (fcNN) to classify images 38

2.2 Convolutional NNs for image-like data 44
Main ideas in a CNN architecture 44 A minimal CNN for edge
■

lovers 47 Biological inspiration for a CNN architecture 50

■

Building and understanding a CNN 52

2.3 One-dimensional CNNs for ordered data 56
Format of time-ordered data 57 What’s special about ordered
■

data? 58 Architectures for time-ordered data 59

■

3 Principles of curve fitting

3.1
62
“Hello world” in curve fitting 63
Fitting a linear regression model based on a loss function 65
3.2 Gradient descent method 69
Loss with one free model parameter 69 ■ Loss with two free
model parameters 73
3.3 Special DL sauce 78
Mini-batch gradient descent 78 Using SGD variants to speed up
■

the learning 79 Automatic differentiation 79

■

3.4 Backpropagation in DL frameworks 80

Static graph frameworks 81 ■ Dynamic graph frameworks 88

PART 2 MAXIMUM LIKELIHOOD APPROACHES FOR

PROBABILISTIC DL MODELS ...............................91

4 Building loss functions with the likelihood approach

4.1 Introduction to the MaxLike principle: The mother
of all loss functions 94
93
 CONTENTS ix

4.2 Deriving a loss function for a classification problem 99

Binary classification problem 99 Classification problems with
■

more than two classes 105 Relationship between NLL, cross

■

entropy, and Kullback-Leibler divergence 109

4.3 Deriving a loss function for regression problems 111
Using a NN without hidden layers and one output neuron for
modeling a linear relationship between input and output 111
Using a NN with hidden layers to model non-linear relationships
between input and output 119 Using an NN with additional
■

output for regression tasks with nonconstant variance 121

5 Probabilistic deep learning models with TensorFlow

Probability 128
5.1 Evaluating and comparing different probabilistic
prediction models 130
5.2 Introducing TensorFlow Probability (TFP) 132
5.3 Modeling continuous data with TFP 135
Fitting and evaluating a linear regression model with constant
variance 136 Fitting and evaluating a linear regression model
■

with a nonconstant standard deviation 140

5.4 Modeling count data with TensorFlow Probability 145
The Poisson distribution for count data 148 Extending the
■

Poisson distribution to a zero-inflated Poisson (ZIP)

distribution 153

6 Probabilistic deep learning models in the wild

6.1 Flexible probability distributions in state-of-the-art
157

DL models 159
Multinomial distribution as a flexible distribution 160
Making sense of discretized logistic mixture 162
6.2 Case study: Bavarian roadkills 165
6.3 Go with the flow: Introduction to normalizing
flows (NFs) 166
The principle idea of NFs 168 The change of variable technique
■

for probabilities 170 Fitting an NF to data 175 Going

■ ■

deeper by chaining flows 177 Transformation between higher

■

dimensional spaces* 181 Using networks to control flows 183

■

Fun with flows: Sampling faces 188

x CONTENTS

PART 3 BAYESIAN APPROACHES FOR PROBABILISTIC

DL MODELS ....................................................195

7 Bayesian learning
7.1
197
What’s wrong with non-Bayesian DL:
The elephant in the room 198
7.2 The first encounter with a Bayesian approach 201
Bayesian model: The hacker’s way 202 ■
What did we
just do? 206
7.3 The Bayesian approach for probabilistic models 207
Training and prediction with a Bayesian model 208 A coin toss ■

as a Hello World example for Bayesian models 213 Revisiting ■

the Bayesian linear regression model 224

8 Bayesian neural networks

8.1
229
Bayesian neural networks (BNNs) 230
8.2 Variational inference (VI) as an approximative
Bayes approach 232
Looking under the hood of VI* 233 ■ Applying VI to the toy
problem* 238
8.3 Variational inference with TensorFlow Probability 243
8.4 MC dropout as an approximate Bayes approach 245
Classical dropout used during training 246 ■
MC dropout used
during train and test times 249
8.5 Case studies 252
Regression case study on extrapolation 252 ■ Classification case
study with novel classes 256

Glossary of terms and abbreviations 264

index 269
 preface
Thank you for buying our book. We hope that it provides you with a look under the
hood of deep learning (DL) and gives you some inspirations on how to use probabilis-
tic DL methods for your work.
All three of us, the authors, have a background in statistics. We started our journey
in DL together in 2014. We got so excited about it that DL is still in the center of our
professional lives. DL has a broad range of applications, but we are especially fasci-
nated by the power of combining DL models with probabilistic approaches as used in
statistics. In our experience, a deep understanding of the potential of probabilistic DL
requires both insight into the underlying methods and practical experience. There-
fore, we tried to find a good balance of both ingredients in this book.
In this book, we aimed to give some clear ideas and examples of applications
before discussing the methods involved. You also have the chance to make practical
use of all discussed methods by working with the accompanying Jupyter notebooks.
We hope you learn as much by reading this book as we learned while writing it. Have
fun and stay curious!

xi
 acknowledgments
We want to thank all the people who helped us in writing this book. A special thanks
go out to our development editor, Marina Michaels, who managed to teach a bunch of
Swiss and Germans how to write sentences shorter than a few hundred words. Without
her, you would have no fun deciphering the text. Also, many thanks to our copyeditor,
Frances Buran, who spotted uncountable errors and inconsistencies in the text (and
also in the formulas, kudos!). We also got much support on the technical side from Al
Krinkler and Hefin Rhys to make the text and code in the notebooks more consistent
and easier to understand. Also, thank you to our project editor, Deirdre Hiam; our
proofreader, Keri Hales; and our review editor, Aleksandar Dragosavljević. We would
also like to thank the reviewers, which at various stages of the book helped with their
very valuable feedback: Bartek Krzyszycha, Brynjar Smári Bjarnason, David Jacobs,
Diego Casella, Francisco José Lacueva Pérez, Gary Bake, Guillaume Alleon, Howard
Bandy, Jon Machtynger, Kim Falk Jorgensen, Kumar Kandasami, Raphael Yan, Richard
Vaughan, Richard Ward, and Zalán Somogyváry.
Finally, we would also like to thank Richard Sheppard for the many excellent
graphics and drawings making the book less dry and friendlier.
I, Oliver, would like to thank my partner Lena Obendiek for her patience as I
worked on the book for many long hours. I also thank my friends from the “Tatort”
viewing club for providing food and company each Sunday at 8:15 pm and for keep-
ing me from going crazy while writing this book.
I, Beate, want to thank my friends, not so much for helping me to write the book,
but for sharing with me a good time beyond the computer screen—first of all my

xii
 ACKNOWLEDGMENTS xiii

partner Michael, but also the infamous Limmat BBQ group and my friends and family
outside of Zurich who still spend leisure time with me despite the Rösti-Graben, the
country border to the big canton, or even the big pond in between.
I, Elvis, want to thank everyone who supported me during the exciting time of writ-
ing this book, not only professionally, but also privately during a good glass of wine or
a game of football.
We, the Tensor Chiefs, are happy that we made it together to the end of this book.
We look forward to new scientific journeys, but also to less stressful times where we not
only meet for work, but also for fun.
 about this book
In this book, we hope to bring the probabilistic principles underpinning deep learn-
ing (DL) to a broader audience. In the end (almost), all neural networks (NNs) in DL
are probabilistic models.
There are two powerful probabilistic principles: maximum likelihood and Bayes.
Maximum likelihood (fondly referred to as MaxLike) governs all traditional DL.
Understanding networks as probabilistic models trained with the maximum likeli-
hood principle helps you to boost the performance of your networks (as Google did
when going from WaveNet to WaveNet++) or to generate astounding applications
(like OpenAI did with Glow, a net that generates realistic looking faces). Bayesian
methods come into play in situations where networks need to say, “I’m not sure.”
(Strangely, traditional NNs cannot do this.) The subtitle for the book, “with Python,
Keras, and TensorFlow Probability,” reflects the fact that you really should get your
hands dirty and do some coding.

Who should read this book

This book is written for people who like to understand the underlying probabilistic
principles of DL. Ideally, you should have some experience with DL or machine learn-
ing (ML) and should not be too afraid of a bit of math and Python code. We did not
spare the math and always included examples in code. We believe math goes better
with code.

xiv
 ABOUT THIS BOOK xv

How this book is organized: A roadmap

The book has three parts that cover eight chapters. Part 1 explains traditional deep
learning (DL) architectures and how the training of neural networks (NNs) is done
technically.
■ Chapter 1—Sets the stage and introduces you to probabilistic DL.
■ Chapter 2—Talks about network architectures. We cover fully connected neural
networks (fcNNs), which are kind of all-purpose networks, and convolutional
neural networks (CNNs), which are ideal for images.
■ Chapter 3—Shows you how NNs manage to fit millions of parameters. We keep
it easy and show gradient descent and backpropagation on the simplest network
one can think of—linear regression.
Part 2 focuses on using NNs as probabilistic models. In contrast to part 3, we discuss
maximum likelihood approaches. These are behind all traditional DL.
■ Chapter 4—Explores maximum likelihood (MaxLike), the underlying princi-
ple of ML and DL. We start by applying this principle to classification and
(simple regression problems).
■ Chapter 5—Introduces TensorFlow Probability (TFP), a framework to build
deep probabilistic models. We use it for not-so-simple regression problems like
count data.
■ Chapter 6—Begins with more complex regression models. At the end, we
explain how you can use probabilistic models to master complex distributions
like describing images of human faces.
Part 3 introduces Bayesian NNs. Bayesian NNs allow you to handle uncertainty.
■ Chapter 7—Motivates the need for Bayesian DL and explains its principles. We
again look at the simple example of linear regression to explain the Bayesian
principle.
■ Chapter 8—Shows you how to build Bayesian NNs. Here we cover two approaches
called MC (Monte Carlo) dropout and variational inference.
If you already have experience with DL, you can skip the first part. Also, the second
part of chapter 6 (starting with section 6.3) describes normalizing flows. You do not
need to know these to understand the material in part 3. Section 6.3.5 is a bit heavy on
math, so if this is not your cup of tea, you can skip it. The same holds true for sec-
tions 8.2.1 and 8.2.2.

About the code

This book contains many examples of source code both in numbered listings and in
line with normal text. In both cases, source code is formatted in a fixed-width font,
like this to separate it from ordinary text.
The code samples are taken from Jupyter notebooks. These notebooks include
additional explanations and most include little exercises you should do for a better
xvi ABOUT THIS BOOK

understanding of the concepts introduced in this book. You can find all the code in
this directory in GitHub: https://github.com/tensorchiefs/dl_book/. A good place to
start is in the directory https://tensorchiefs.github.io/dl_book/, where you’ll find
links to the notebooks. The notebooks are numbered according to the chapters. So,
for example, nb_ch08_02 is the second notebook in chapter 8.
All the examples in this book, except nb_06_05, are tested with the TensorFlow v2.1
and TensorFlow Probability (TFP) v0.8. The notebooks nb_ch03_03 and nb_ch03_04,
describing the computation graphs, are easier to understand in TensorFlow v1. For
these notebooks, we also include both versions of TensorFlow. The nb_06_05 note-
book only works with TensorFlow v1 because we need weights that are only provided
in that version of TensorFlow.
You can execute the notebooks in Google’s Colab or locally. Colab is great; you can
simply click on a link and then play with the code in the cloud. No installation—you
just need a browser. We definitely suggest that you go this way.
TensorFlow is still fast-evolving, and we cannot guarantee the code will run in sev-
eral years’ time. We, therefore, provide a Docker container (https://github.com oduerr/
dl_book_docker/) that you can use to execute all notebooks except nb_06_05 and the
TensorFlow 1.0 versions of nb_ch03_03 and nb_ch03_04. This Docker container is
the way to go if you want to use the notebooks locally.

liveBook discussion forum

Purchase of Probabilistic Deep Learning includes free access to a private web forum run
by Manning Publications where you can make comments about the book, ask techni-
cal questions, and receive help from the authors and from other users. To access the
forum, go to https://livebook.manning.com/book/probabilistic-deep-learning-with-
python/welcome/v-6/. You can also learn more about Manning’s forums and the
rules of conduct at https://livebook.manning.com/#!/discussion.
Manning’s commitment to our readers is to provide a venue where a meaningful
dialogue between individual readers and between readers and the authors can take
place. It is not a commitment to any specific amount of participation on the part of
the authors, whose contribution to the forum remains voluntary (and unpaid). We
suggest you try asking the authors some challenging questions lest their interest stray!
The forum and the archives of previous discussions will be accessible from the pub-
lisher’s website as long as the book is in print.
 about the authors
Oliver Dürr is professor for data science at the University of Applied Sciences in Kon-
stanz, Germany. Beate Sick holds a chair for applied statistics at ZHAW, and works as a
researcher and lecturer at the University of Zurich, and as a lecturer at ETH Zurich.
Elvis Murina is a research scientist, responsible for the extensive exercises that accom-
pany this book.
Dürr and Sick are both experts in machine learning and statistics. They have
supervised numerous bachelor’s, master’s, and PhD theses on the topic of deep learn-
ing, and planned and conducted several postgraduate- and master’s-level deep learning
courses. All three authors have worked with deep learning methods since 2013, and
have extensive experience in both teaching the topic and developing probabilistic deep
learning models.

xvii
 about the cover illustration
The figure on the cover of Probabilistic Deep Learning is captioned “Danseuse de l’Isle
O-tahiti,” or A dancer from the island of Tahiti. The illustration is taken from a collec-
tion of dress costumes from various countries by Jacques Grasset de Saint-Sauveur
(1757–1810), titled Costumes de Différents Pays, published in France in 1788. Each illustra-
tion is finely drawn and colored by hand. The rich variety of Grasset de Saint-Sauveur’s
collection reminds us vividly of how culturally apart the world’s towns and regions
were just 200 years ago. Isolated from each other, people spoke different dialects and
languages. In the streets or in the countryside, it was easy to identify where they lived
and what their trade or station in life was just by their dress.
The way we dress has changed since then and the diversity by region, so rich at the
time, has faded away. It is now hard to tell apart the inhabitants of different conti-
nents, let alone different towns, regions, or countries. Perhaps we have traded cultural
diversity for a more varied personal life—certainly for a more varied and fast-paced
technological life.
At a time when it is hard to tell one computer book from another, Manning cele-
brates the inventiveness and initiative of the computer business with book covers
based on the rich diversity of regional life of two centuries ago, brought back to life by
Grasset de Saint-Sauveur’s pictures.

xviii
 Part 1

Basics of deep learning

P art 1 of this book gives you a first high-level understanding of what probabi-
listic deep learning (DL) is about and which types of tasks you can tackle with it.
You’ll learn about different neural network architectures for regression (that
you can use to predict a number), and about classification (that you can use to
predict a class). You’ll get practical experiences in setting up DL models, learn
how to tune these, and learn how to control the training procedure. If you don’t
already have substantial experience with DL, you should work through part 1 in
full before moving on to the probabilistic DL models in part 2.
 Introduction to
probabilistic deep learning

Let’s use my
probabilistic travel
You’ll get $500 time gadget!
tip if I arrive at
MoMA within
25 minutes!

This chapter covers

 What is a probabilistic model?
 What is deep learning and when do you use it?
 Comparing traditional machine learning and deep
learning approaches for image classification
 The underlying principles of both curve fitting
and neural networks
 Comparing non-probabilistic and probabilistic
models
 What probabilistic deep learning is and why
it’s useful

3
4 CHAPTER 1 Introduction to probabilistic deep learning

Deep learning (DL) is one of the hottest topics in data science and artificial intelli-
gence today. DL has only been feasible since 2012 with the widespread usage of GPUs,
but you’re probably already dealing with DL technologies in various areas of your
daily life. When you vocally communicate with a digital assistant, when you translate
text from one language into another using the free DeepL translator service (DeepL
is a company producing translation engines based on DL), or when you use a search
engine such as Google, DL is doing its magic behind the scenes. Many state-of-the-art
DL applications such as text-to-speech translations boost their performance using
probabilistic DL models. Further, safety critical applications like self-driving cars use
Bayesian variants of probabilistic DL.
In this chapter, you will get a first high-level introduction to DL and its probabilis-
tic variants. We use simple examples to discuss the differences between non-probabilistic
and probabilistic models and then highlight some advantages of probabilistic DL mod-
els. We also give you a first impression of what you gain when working with Bayesian
variants of probabilistic DL models. In the remaining chapters of the book, you will
learn how to implement DL models and how to tweak them to get their more power-
ful probabilistic variants. You will also learn about the underlying principles that
enable you to build your own models and to understand advanced modern models so
that you can adapt them for your own purposes.

1.1 A first look at probabilistic models

Let’s first get an idea of what a probabilistic model can look like and how you can use
it. We use an example from daily life to discuss the difference between a non-probabilis-
tic model and a probabilistic model. We then use the same example to highlight some
advantages of a probabilistic model.
In our cars, most of us use a satellite navigational system (satnav—a.k.a. GPS) that
tells us how to get from A to B. For each suggested route, the satnav also predicts the
needed travel time. Such a predicted travel time can be understood as a best guess.
You know you’ll sometimes need more time and sometimes less time when taking the
same route from A to B. But a standard satnav is non-probabilistic: it predicts only a
single value for the travel time and does not tell you a possible range of values. For an
example, look at the left panel in figure 1.1, where you see two routes going from
Croxton, New York, to the Museum of Modern Art (MoMA), also in New York, with a
predicted travel time that is the satnav’s best guess based on previous data and the cur-
rent road conditions.
Let’s imagine a fancier satnav that uses a probabilistic model. It not only gives you
a best guess for the travel time, but also captures the uncertainty of that travel time.
The probabilistic prediction of the travel time for a given route is provided as a distri-
bution. For example, look at the right panel of figure 1.1. You see two Gaussian bell
curves describing the predicted travel-time distributions for the two routes.
How can you benefit from knowing these distributions of the predicted travel
time? Imagine you are a New York cab driver. At Croxton, an art dealer boards your
 A first look at probabilistic models 5

taxi. She wants to participate in a great art auction that starts in 25 minutes and offers
you a generous tip ($500) if she arrives there on time. That’s quite an incentive!
Your satnav tool proposes two routes (see the left panel of figure 1.1). As a first
impulse, you would probably choose the upper route because, for this route, it esti-
mates a travel time of 19 minutes, which is shorter than the 22 minutes for the other
route. But, fortunately, you always have the newest gadgets, and your satnav uses a
probabilistic model that not only outputs the mean travel time but also a whole distri-
bution of travel times. Even better, you know how to make use of the outputted distri-
bution for the travel times.
You realize that in your current situation, the mean travel time is not very inter-
esting. What really matters to you is the following question: With which route do you

Let’s use my
probabilistic travel
You’ll get $500 time gadget!
tip if I arrive at
MoMA within
25 minutes!

Chance to get tip: 69%

0.03
Density
0.02

19 Min.
7.9 Miles
0.01

0 10 20 30 40
Travel time

Chance to get tip: 93%

0.02

22 Min.
8.2 Miles
Density
0.01
0.00

0 10 20 30 40
Travel time

Figure 1.1 Travel time prediction of the satnav. On the left side of the map, you see a deterministic version—
just a single number is reported. On the right side, you see the probability distributions for the travel time of the
two routes.
6 CHAPTER 1 Introduction to probabilistic deep learning

have the better chance of getting the $500 tip? To answer this question, you can look
at the distributions on the right side of figure 1.1. After a quick eyeball analysis, you
conclude that you have a better chance of getting the tip when taking the lower
route, even though it has a larger mean travel time. The reason is that the narrow
distribution of the lower route has a larger fraction of the distribution correspond-
ing to travel times shorter than 25 minutes. To support your assessment with hard num-
bers, you can use the satnav tool with the probabilistic model to compute for both
distributions the probability of arriving at MoMA in less than 25 minutes. This probabil-
ity corresponds to the proportion of the area under the curve left of the dashed line in
figure 1.1, which indicates a critical value of 25 minutes. Letting the tool compute the
probabilities from the distribution, you know that your chance of getting the tip is 93%
when taking the lower route and only 69% when taking the upper road.
As discussed in this cab driver example, the main advantages of probabilistic mod-
els are that these can capture the uncertainties in most real-world applications and
provide essential information for decision making. Other examples of the use of prob-
abilistic models include self-driving cars or digital medicine probabilistic models. You
can also use probabilistic DL to generate new data that is similar to your observed
data. A famous fun application is to create realistic looking faces of non-existing peo-
ple. We talk about this in chapter 6. Let’s first look at DL from a bird’s-eye view before
peeking into the curve-fitting part.

1.2 A first brief look at deep learning (DL)

What is DL anyway? When asked for a short elevator pitch, we would say that it’s a
machine learning (ML) technique based on artificial neural networks (NNs) and that it’s
loosely inspired by the way the human brain works. Before giving our personal definition
of DL, we first want to give you an idea of what an artificial NN looks like (see figure 1.2).

Hidden Hidden Hidden

Input layer 1 layer 2 layer 3
layer

Output
layer

Figure 1.2 An example of an artificial neural network (NN) model with three hidden layers. The input
layers hold as many neurons as we have numbers to describe the input.
 A first brief look at deep learning (DL) 7

In figure 1.2, you can see a typical traditional artificial NN with three hidden layers
and several neurons in each layer. Each neuron within a layer is connected with each
neuron in the next layer.
An artificial NN is inspired by the brain that consists of up to billions of neurons
processing, for example, all sensory perceptions such as vision or hearing. Neurons
within the brain aren’t connected to every other neuron, and a signal is processed
through a hierarchical network of neurons. You can see a similar hierarchical network
structure in the artificial NN shown in figure 1.2. While a biological neuron is quite
complex in how it processes information, a neuron in an artificial NN is a simplifica-
tion and abstraction of its biological counterpart.
To get a first idea about an artificial NN, you can better imagine a neuron as a con-
tainer for a number. The neurons in the input layer are correspondingly holding the
numbers of the input data. Such input data could, for example, be the age (in years),
income (in dollars), and height (in inches) of a customer. All neurons in the following
layers get the weighted sum of the values from the connected neurons in the previous
layer as their input. In general, the different connections aren’t equally important but
have weights, which determine the influence of the incoming neuron’s value on the
neuron’s value in the next layer. (Here we omit that this input is further transformed
within the neuron.) DL models are NNs, but they also have a large number of hidden
layers (not just three as in the example from figure 1.2).
The weights (strength of connections between neurons) in an artificial NN need
to be learned for the task at hand. For that learning step, you use training data and
tune the weights to optimally fit the data. This step is called fitting. Only after the fit-
ting step can you use the model to do predictions on new data.
Setting up a DL system is always a two-stage process. In the first step, you choose an
architecture. In figure 1.2, we chose a network with three layers in which each neuron
from a given layer is connected to each neuron in the next layer. Other types of net-
works have different connections, but the principle stays the same. In the next step,
you tune the weights of the model so that the training data is best described. This fit-
ting step is usually done using a procedure called gradient descent. You’ll learn more
about gradient descent in chapter 3.
Note that this two-step procedure is nothing special to DL but is also present in
standard statistical modeling and ML. The underlying principles of fitting are the
same for DL, ML, and statistics. We’re convinced that you can profit a lot by using the
knowledge that was gained in the field of statistics during the last centuries. This book
acknowledges the heritage of traditional statistics and builds on it. Because of this, you
can understand much of DL by looking at something as simple as linear regression,
which we introduce in this chapter and use throughout the book as an easy example.
You’ll see in chapter 4 that linear regression already is a probabilistic model providing
more information than just one predicted output value for each sample. In that chapter,
you’ll learn how to pick an appropriate distribution to model the variability of the
outcome values. In chapter 5, we’ll show you how to use the TensorFlow Probability
8 CHAPTER 1 Introduction to probabilistic deep learning

framework to fit such a probabilistic DL model. You can then transfer this approach to
new situations allowing you to design and fit appropriate probabilistic DL models that
not only provide high performance predictions but also capture the noise of the data.

1.2.1 A success story

DL has revolutionized areas that so far have been especially hard to master with tradi-
tional ML approaches but that are easy to solve by humans, such as the ability to recog-
nize objects in images (computer vision) and to process written text (natural language
processing) or, more generally, any kind of perception tasks. Image classification is far
from being only an academic problem and is used for a variety of applications:
 Face recognition
 Diagnostics of brain tumors in MRI data
 Recognition of road signs for self-driving cars

Although DL reveals its potential in different application areas, probably the easiest to
grasp is in the field of computer vision. We therefore use computer vision to motivate
DL by one of its biggest success stories.
In 2012, DL made a splash when Alex Krizhevsky from Geoffrey Hinton’s lab
crushed all competitors in the internationally renowned ImageNet competition with a
DL-based model. In this competition, teams from leading computer vision labs
trained their models on a big data set of ~1 million images with the goal of teaching
these to distinguish 1,000 different classes of image content. Examples for such classes
are ships, mushrooms, and leopards. In the competition, all trained models had to list
the five most probable classes for a set of new test images. If the right class wasn’t
among the proposed classes, the test image counted as an error (see figure 1.3, which
shows how DL-based approaches took image classification by storm).
Before DL entered the competition, the best programs had an error rate of ~25%.
In 2012, Krizhevsky was the first to use DL and achieved a huge drop in the error rate
(by 10% to only ~15%). Only a year later, in 2013, almost all competitors used DL, and
in 2015, different DL-based models reached the level of human performance, which is
about 5%. You might wonder why humans misclassify 1 image in 20 (5%). A fun fact:
there are 170 different dog breeds in that data set, which makes it a bit harder for
humans to correctly classify the images.

1.3 Classification
Let’s look at the differences between non-probabilistic, probabilistic, and Bayesian
probabilistic classification. DL is known to outperform traditional methods, espe-
cially in image classification tasks. Before going into details, we want to use a face
recognition problem to give you a feeling for the differences and the commonalities
between a DL approach and a more traditional approach to face recognition. As a
side note, face recognition is actually the application that initially brought us into
contact with DL.
 Classification 9

79%
Traditional CV Deep Learning Human: 5% misclassification
70%

60%

50%
Error Rate

40%
Only one non-CNN
30% approach in 2013

20%

10%
GoogleNet 6.7%
7%

2010 2011 2012 2013 2014

A. Krizhevsky 2015: It gets tougher

first CNN in 2012 4.95% Microsoft (Feb 6 surpassing human performance 5.1%)
4.8% Google (Feb 11), further improved to 3.6 (Dec)
4.58% Baidu (May 11 banned due too many submissions)
3.57% Microsoft (Resnet winner 2015)

Figure 1.3 The impressive results of DL in the ImageNet competition

As statisticians, we had a collaboration project with some computer science colleagues

for doing face recognition on a Raspberry Pi minicomputer. The computer scientists
challenged us by kidding about the age of the used statistical methods. We took the
challenge and brought them to a surprised silence by proposing DL to tackle our face
recognition problem. The success in this first project triggered many other joint DL
projects, and our interests grew, looking deeper into the underlying principles of
these models.
Let’s look at a specific task. Sara and Chantal were together on holidays and took
many pictures, each showing at least one of them. The task is to create a program that
can look at a photo and determine which of the two women is in the photo. To get a
training data set, we labeled 900 pictures, 450 for each woman, with the name of the
pictured woman. You can imagine that images can be very different at first sight
because the women might be pictured from different angles, laughing or tired,
dressed up or casual, or having a bad hair day. Still, for you, the task is quite easy. But
for a computer, an image is only an array of pixel values, and programming it to tell
the difference between two women is far from trivial.

1.3.1 Traditional approach to image classification

A traditional approach to image classification doesn’t directly start with the pixel val-
ues of the images but tackles the classification task in a two-step process. As a first step,
experts in the field define features that are useful to classify the images. A simple
example of such a feature would be the mean intensity value of all pixels, which can
Another Random Scribd Document
with Unrelated Content
 This crop is closely related to cabbage and cauliflower. Instead of a single
head, Brussels sprouts form a large number of small heads in the axils of the
leaves. As the heads begin to crowd, the leaves should be broken from the stem
of the plant to give them more room. A few leaves should be left at the top of
the stem where the new heads are being formed. Brussels sprouts are more
hardy than cabbage, and in mild climates may remain in the open ground all
winter, the heads being removed as desired. For winter use in cold localities, take
up plants that are well laden with heads and set them close together in a pit,
cold frame, or cellar, with a little soil around the roots. The uses of Brussels
sprouts are similar to those of cabbage, but they are considered to be of a
superior flavor. They require the same treatment as cabbage. The soil must be
rich and requires considerable moisture. The small sprouts must grow rapidly or
they will be tough. Sow the seed in hotbed and transplant, or scatter seed in hills
and thin. The plants must have plenty of room. Rows should be thirty inches
apart and the plants not closer than two feet.—(F. B. 255; U. Idaho E. S. 10;
Cornell U. E. S. 292.)

CABBAGE.
Cabbage is one of the most universally cultivated of the garden plants.
Although it is one of the coarser vegetables it finds a place in the home garden
as well as in the market garden and truck farm. In some sections of the United
States it is extensively grown as a farm crop. Early cabbage is practically all
consumed as a green vegetable. The late crop, on the other hand, is handled as
a fresh vegetable, as a storage crop, and for the manufacture of sauerkraut. It is
always in demand, and under present conditions is always available, either as the
product of a southern truck farm or a northern farm, garden, or storage house.
The group of cultivated plants which has been derived from the wild cabbage
presents a greater diversity of form than that derived from any other single
ancestral type.
Wild cabbage is a robust-growing broad-leaved plant enjoying the low, moist
areas near the seacoast of southern Europe. The most closely allied form now in
cultivation is the collard. The wide variation in the group is illustrated by the
diversity of form shown in collards, kale, tree cabbage, marrow kale, cauliflower,
and Brussels sprouts. It is almost beyond the bounds of reason to believe that all
these forms have been derived from a common parentage, yet such is the fact.
Seed.—In no truck crop does the character of the seed count for more than in
cabbage. It is very essential that the crop come to marketable maturity early,
that the heads be uniform in size and character, and that they mature so that the
whole crop can be harvested at two cuttings. The small saving made by the
purchase of cheap or inferior seed is usually paid for a hundred times over in the
lessened value of the crop. A grower can not afford to risk his crop for so small a
saving. The best seed that can be obtained is none too good, and anything short
of this is not good business. Without highly viable seed of a good strain, true to
type, the best results can not be expected. For early spring cabbage in the
South, sow the seeds in an outdoor bed and transplant to the garden before
January 1. In the North, plant the seeds in a hotbed during February and set the
plants in the open ground as early as the soil can be worked. For a late crop in
the North, plant the seeds in a bed in the open ground in May or June and
transplant to the garden in July. Early cabbages require a rich, warm soil in order
that they may mature early. For late cabbages the soil should be heavier and
more retentive of moisture and not so rich as for the early crop, as the heads are
liable to burst. Cabbages should be set in rows 30 to 36 inches apart and 14 to
18 inches apart in the row. Where the plants are set out in the autumn and
allowed to remain in the ground over winter, they are usually placed on top of
ridges.
Soil.—The soil for cabbage must necessarily vary in different localities. In one
area it may be of an alluvial character, while in another it may be sedentary, and
in still another it may be characteristic glacial drift. The fact that cabbage grows
well in all these soils indicates its adaptation to a wide range of conditions. The
main thing with cabbage is an abundant supply of immediately available plant
food. Market gardeners rely chiefly upon stable manure for their supply of plant
food.
Cultivation.—Among market gardeners it is a common expression that
"cabbage should be hoed every day." Perhaps no other crop responds more
quickly to good cultivation and an ample food supply. This is undoubtedly the
explanation of the above quoted expression. In cultivating cabbage the work
should be frequent and thorough, but the cultivation should not be deep. The
aim should be to destroy all competing weeds and to maintain a loose, friable
layer of soil about 2 inches deep over the surface of the area devoted to
cabbage.
Storage.—Early cabbage must be used soon after it has formed solid heads,
as it will not keep during hot weather. Late cabbage may be buried in pits or
stored in cellars or specially constructed houses. The usual method of storing
cabbage is to dig a trench about 18 inches deep and 3 feet wide and set the
cabbage upright, with the heads close together and the roots bedded in soil. As
cold weather comes on, the heads are covered slightly with straw and then 3 or
4 inches of earth put on. Slight freezing does not injure cabbage, but it should
not be subjected to repeated freezing and thawing. If stored in a cellar or
building, the heads are generally cut from the stems and stored on slatted
shelves or in shallow bins. While in storage, cabbage should be well ventilated
and kept as cool as possible without freezing.
Varieties.—The varieties of cabbage used in the trucking section are
practically limited to the Wakefield type. There are two strains of this type of
cabbage now extensively employed: The true Jersey Wakefield, with its small,
acutely pointed tip and very firm, tender flesh of high quality, and the Charleston
Wakefield, which is broader, somewhat flatter, more obtusely pointed, and
slightly more angular in cross section than the Jersey type. The varieties which
may be used for field cultivation depend upon the purpose for which the cabbage
is intended. If for sauerkraut or for immediate consumption, the Flat Dutch type
from American-grown seed is extensively employed in the eastern part of the
United States. In the irrigated section of Colorado, in the vicinity of Greeley,
where cabbage is grown for sauerkraut, a variety known as Scotch Cross is
almost universally grown. If the cabbage is intended for storage the Danish Ball
Head from imported seed is almost exclusively used.—(F. B. 255, 433; Colo. E. S.
143; Md. Ag. Col. E. S. 133; Tex. E. S. 52, 69; Ga. E. S. 91; Kans. E. S. 70; S.
Dak. E. S. 91.)

CALABASH[1]
The increasing popularity of calabash pipes made from the fruits of a South
African calabash, or gourd, has aroused a widespread interest in the growing of
this vine.
Calabash pipes made from imported South African gourds have been the
fashion in England for some time and are now coming into vogue in America.
These pipes are formed from the crooked necks of a large gourd (Lagenaria
vulgaris) belonging to the well-known group of plants which includes the
cucumber, the melons, and the squashes. Pipes made from the imported gourds
are expensive, American dealers usually charging $3 to $12 apiece for them.
They are the lightest pipes made for their size, are graceful in shape, color like
meerschaums, and are delightful smokers. Unlike the cheap pipes which are
turned out by machinery, no two of these calabash pipes are alike. In this lies
much of their charm. In this, likewise, lies their cost, for, unlike the great mass of
pipes turned out by machinery, the crook of the calabash varies so that each
mouthpiece must be made to fit it and each lining of meerschaum or plaster of
Paris must be specially adapted. In our land of labor-saving machinery and
expensive hand labor this is what makes the pipes costly.
The vine forms a very satisfactory cover for unsightly brush heaps or fences,
though its rather rank odor might prove objectionable if used for an arbor too
near the dwelling. To grow the vine for the sake of its gourds is where the chief
interest lies, however, and to do this well it should not be trained on a trellis, but
allowed to trail over the ground. If the fruits are allowed to lie on the ground
they form their crooked necks quite naturally without assistance, and while not
all of them by any means make suitable necks for pipes a good proportion do. It
seems to induce a more perfect neck to stand the gourds up when half grown so
that they rest on their big ends. Unless care is exercised in doing this the necks
snap off, for they are extremely brittle even when fully grown. It is only when
almost mature that they become hard and then they are indeed almost
unbreakable.
Much could doubtless be done to perfect the methods of culture, insuring
perhaps a greater percentage of properly crooked necks and more perfect
surfaces. It could not be seen that inheritance plays any material part in this
matter of percentage of crooks. If left to themselves the majority will crook their
necks, but some few will remain quite straight, and this on the same vine with
perfectly formed crooks. The gourds should be left as long as possible on the
vines to thoroughly thicken their shells. If picked green the shell will be no
thicker than stiff cardboard and in drying it is very liable to crack. Frost will injure
the gourds if they are left on the vines too long.—(B. P. I. Cir. 41.)

CANTALOUPE.
Cantaloupe growing, as developed since its origin near Rockyford, Colorado,
in 1885, requires unusual judgment and cultural skill on the part of the farmer.
Co-operative organization and good business management are also essential, for
only by these means can the crop be properly timed and prepared for shipment,
and necessary arrangements made with transportation and selling agencies.
Seed.—Seed should be most carefully selected with reference to flavor and
appearance of the fruit; to good shipping characters, including small cavities and
heavy netting; and to a tendency to produce melons of standard size. Early
strains are desired for some situations; but in Arizona rust resistance is not a
necessary character as this class of diseases is little to be feared under the arid
conditions. Seed should be purchased only from most reliable sources. Rockyford
growers are at present the principal means of supply.
Soil.—Experience has proven that a sandy loam is the soil best suited for
cantaloupes, and that its condition of tilth and the available fertility are the prime
essentials in bringing cantaloupes to quick maturity. The secret of getting soil in
that ashy, mellow condition so desirable for cantaloupes is largely one of
experience, for hardly two farms can be handled the same. In general, there
must be moisture in the soil over winter to get the disintegrating effect of frost,
and plowing should not be done until the ground is dry enough to pulverize
mellow. Barnyard manure has long been the means of supplying fertility to force
cantaloupes to early maturity. Old alfalfa ground is most excellent for cantaloupe
culture. Bermuda sod plowed up and exposed to the sun without irrigation the
preceding summer makes excellent cantaloupe ground, the intensive cultivation
necessary serving both to benefit the crop and to restrain this formidable weed.
Planting.—The first requisite aside from moisture for a good start is warm
weather, as cantaloupe seed cannot germinate when the ground is cold and
freezing; and if perchance the days are warm enough to germinate the seed that
is planted in March or April, the cold nights that are sure to follow will offset the
advantage of early planting. If there is a secret in getting early cantaloupes it is
in growing the crop from start to finish with a uniform unchecked growth; the
cantaloupe does not seem to have the power to rally from a check in growth or
an injury from an insect and still makes its normal development. The back-set
not only cuts off the production of early cantaloupes but seriously affects the size
and quality of the melon. There are numerous instances where unfavorable
conditions of growth have produced a large quantity of pony melons, while under
more favorably growing conditions the same seed and soil have yielded standard
sized cantaloupes. One of the first signs of promise for early cantaloupes is a
quick germination and rapid development of large cotyledons. Seed that
germinates slowly with small, yellow appearing seed leaves has never made early
cantaloupes.
Irrigation.—Moisture for the cantaloupe hill is generally supplied by the
irrigation furrow. It should always reach the seed or plant by soaking through the
soil. Irrigation should never be allowed to over-soak or flood the ground, as the
soil will then become hard and not permit a good growth. The relation of
irrigation to an early set of cantaloupes is a somewhat mooted question. There
are growers who argue the use of frequent irrigations during the setting period
to secure a good set, and there are others who prefer to keep the vines rather
dry and even letting them show the need of water before they will irrigate during
the setting stage. There have been results that seemed to support both theories,
yet close observation would not warrant following either plan to an extreme, but
rather a medium course of supplying enough moisture for an even, healthy
growth, which seems to be the essential condition all the way through. An excess
of irrigation during the hot weather in July will doubtless tend to grow vines at
the expense of early fruit; but the most disastrous result of too much water—
having the ground so soaked that the surface is nearly all wet, and affording the
moist, dewy condition which is favorable to its development—is in the
development of rust. The rust problem is a serious one in cantaloupe culture in
Colorado. Controlling it by proper application of irrigation is only a palliative
measure, yet a marked contrast is often seen in two portions of a field; one over-
irrigated, and the other comparatively dry, aside from the moisture necessary to
the growth of the vines. Rainy weather and dewy nights afford the proper
conditions for the growth of the rust spore, and while the farmer cannot change
climatic conditions, yet by careful attention in the application of water, having the
rows well ditched, and with adequate waste laterals to prevent over-soaking and
flooding, the surface of the ground will dry rapidly after a rain or an irrigation.
Thus the dews at night will be less, and in a measure alleviate the effects of rust.
—(U. Ariz. Cir. 77; Ag. Col. Colo. 62, 85, 95 and 108.)

CARDOON.
The cardoon is a thistle-like plant, very similar in appearance to the Globe
artichoke, but is grown as an annual. The seeds are sown in early spring in a
hotbed or cold frame and the plants transplanted later to the open ground. The
cardoon should be planted in rows 3 feet apart and 18 inches apart in the row on
rich soil, where it can secure plenty of moisture and make rapid growth. Toward
autumn the leaves are drawn together and the center blanched in the same
manner as endive. If intended for winter use, the leaves are not blanched in the
garden, but the plants are lifted with considerable earth adhering to the roots
and stored closely in a dark pit or cellar to blanch. The blanched leaf stems are
used for making salads, soups, and stews.—(F. B. 255.)

CARROT.
The culture of the carrot is practically the same as the parsnip, except that
carrots are not thinned so much and are allowed to grow almost as thickly as
planted. Carrots should be dug in the autumn and stored the same as parsnips or
turnips. Any surplus can be fed sparingly to horses, mules or cattle. The roots of
the carrot are used at all times of the year, mostly in soups, but they may be
boiled and served with butter or creamed. Carrots are planted in rows 16 inches
apart and the plants thinned out to 4 inches in the row. Chantenay is an
excellent table carrot of medium size and dark orange color, slightly tapering and
abruptly terminating with a short, fine taproot. The flesh is orange colored,
brittle, juicy and mild flavored. What it lacks in size it makes up in quality and
good shape. Scarlet Intermediate, somewhat larger than Chantenay, is of good
size for table use. In shape more tapering and with a longer taproot. It is dark
orange colored; flavor and quality good. Flesh is quite brittle and orange colored
with a white center. To these two are added two varieties principally grown for
stock feed, similar varieties being grown for table use in many parts of Europe,
and more especially those of the White Belgian variety. Both varieties are of
slender shape, 1½ to 2 inches in diameter, holding their size well, although
averaging 12 inches in length, 3 to 4 inches of which grows above ground and
which as a consequence is colored light green on the outside. White Belgian is
the sweeter of the two, and while the flesh is somewhat coarse, the flavor of it,
when well stewed and mashed, is sweeter and not unlike that of the parsnip.
Victoria, the other variety, is of the same texture, fairly sweet and with a more
pronounced carrot flavor, the flesh instead of white, being light orange colored.
This vegetable can be grown to perfection in Porto Rico almost any time of the
year. It prefers a rich loam and grows very well on a heavy clay which is not too
wet, but a light sandy soil is not well adapted to it. For fertilizer, stable manure
will do when nothing else is available, but a commercial fertilizer, rich in potash
and phosphoric acid, is much to be preferred for this crop.—F. B. 255, 295; Mich.
E. S. 20; N. C. E. S. 132; U. Idaho E. S. 10; P. R. A. E. S. 7.

CAULIFLOWER.
This plant requires a very rich, moist soil. Land that will produce only a fair
crop of cabbage is unfit for cauliflower. If the land is very rich and well fertilized
it may be reasonably expected that the returns from the crop of cauliflower will
more than repay the cost of putting the land in good condition.
Seed.—No more important element enters into the success of the cauliflower
crop than the quality of the seed and to the seed alone is often due the
difference between success and failure, profit and loss. The best seed that can
be secured is the cheapest at any reasonable price, and it should always be
obtained from a well-known, reputable seedsman.
Seed-bed.—This should be carefully prepared. The soil should be enriched
with a liberal application of commercial fertilizer, or thoroughly decomposed
stable manure. After the fertilizer is applied it should be thoroughly worked in to
a depth of three or four inches. From a few days to two weeks should elapse
before the seed is sown for there is great danger in planting seed too soon after
applying commercial fertilizer as the seed is likely to be destroyed by the action
of the mineral substance unless it has been dissolved and thoroughly
incorporated with the soil. The time between the application of the fertilizer and
the sowing of the seed will depend upon the amount of rainfall and it is often
better to wet down the seed-bed each day for four or five days before planting
and not to depend upon the uncertain rainfall. The rows should be about three
inches apart. In six or seven days the young plants should begin to appear and
the ground between the drills should be cultivated. Do not allow the soil to dry
out as the cauliflower plant from seed to head should never be checked. Neither
should the bed be kept too wet, else there is danger of "damping off." The bed
should be carefully watched and if the disease does break out it may be checked
by removing the diseased plants, working the soil, scattering dry sand and
sulphur along the rows and withholding water until the surface soil becomes dry.
It might be pointed out here that about six months must be allowed from the
sowing of the seed until the crop matures.
Transplanting.—The plants should not be allowed to remain long in the seed
rows. If left too long they will soon crowd and become weak and spindling. When
they have reached the height of one inch, they should be pricked off and set in
another portion of the bed. They may be set in rows four inches apart with the
plants one and a half to two inches apart in the rows. Here they should remain
until ready for the field. If care has been exercised all the way through, the
plants will be short, stocky and vigorous. By the time they are four or five inches
high or when the leaves have lapped they are ready for the field. It is not best to
let them get too large, because there is often a delay of a few days in order to
obtain good climatic conditions for setting out. If left too long in the seed bed,
greater care must be exercised in transplanting, else the plants may suffer a
severe check and will button or break irregularly instead of forming smooth well
shaped heads.
Soils and Preparation.—Work should be started on the ground at least a
month before the plants are set out. The cauliflower is a deep rooted plant,
consequently the soil should be prepared deeply. It is not advisable to turn under
the good surface soil and to obviate this ground may be plowed shallow and then
stirred and opened with a bull-tongue to a depth of seven or eight inches. After
this the surface should be cultivated to a depth of two or three inches. Give
thorough preparation by frequent cultivation before the fertilizer is applied,
preparatory to setting out the plants.
Setting Out.—It is best that the plants be set out either just before or
immediately after a rain, but if this can not be done they should be set out late in
the evening and watered, giving each plant about a quart of water. A cloudy day
is much preferable to a clear one and if the day on which the plants are set out is
followed by cloudy weather, so much the better. The ground should be leveled or
smoothed over, for which purpose a roller or float may be used. After this the
ground may be marked off. Two markers should be constructed, one with the
teeth three feet apart, the other with the teeth two feet apart. These may be
made of wood after the pattern of an ordinary garden rake. In place of a marker
a line may be used or the ground may be checked off with a light hand plow.
Only a limited number of plants should be removed from the seed-bed at one
time. The leaves should be cut back about one-half or one-third, using for the
purpose a large pair of shears. Sprinkle the plants with water as soon as
removed from the bed, place in a shallow box or basket and keep them shaded
from the sun.
 Cultivation and Care.—The field should be frequently cultivated and the
ground should be scarified at least every week and after every rainfall. The best
tool for cultivating is an ordinary cultivator and the ground should not be worked
to a greater depth than two and one-half or three inches. This will preserve a
surface mulch of dry earth and prevent loss of moisture by evaporation. As soon
as the heads commence to form the leaves should be drawn together at the top
and loosely tied near their tips with a piece of cord or twine. Rafia makes a good
substitute for twine and is preferable because there is less danger of cutting the
leaves. The practice of breaking down the leaves over the head has been tried,
but found not quite so satisfactory. If the heads are left uncovered they become
yellow through the action of the sun and rain but when the leaves are drawn
together and tied, they bleach out pure white, and curd-like.
Gathering.—Cauliflower may be cut before it is mature, but the flavor is not
so well developed as it is when the heads are full grown. For winter shipment
heads from four to six inches in diameter are of a desirable size and the market
will take them fully as well or better than large ones. The field should be picked
over at least every two or three days during the season, though heads will
remain in good condition for nearly a week if the weather be cold. Examine the
head by separating the leaves on the side. As soon as the head is well rounded
up in the center and developed so as to force the leaves outward, and assumes a
grained appearance, it will be found to be fully matured. The heads should be
cut, preferably, when dry. If moist they are likely to decay in transit. The best
time of day is the afternoon if they are intended for long distance shipment.
About an inch of stem should be left on the head and three rows of leaves. After
cutting, the heads should be carefully placed in a wagon and carried to the
packing house or on dry pleasant days packing may be done in the field.
Packing.—The package recommended for general use is the ordinary lettuce
basket. Before packing, the leaves should be cut back to stubs. Each head should
be carefully wrapped in a large sheet of white glazed paper. The baskets should
be packed snug and tight without bruising the heads, and only those of uniform
size should be placed in each basket. Never place different sizes in the same
package and always discard inferior or injured heads; the compost heap is the
place for them.—(F. B. 255; Fla. E. S. 59; Tex. E. S. 57; Cornell U. E. S. 292.)

CELERIAC.
This vegetable, which is also known as turnip-rooted celery, or knot celery, is
closely related to our ordinary celery, being indeed a cultural variety of the same
original plant grown under conditions which have developed the root rather than
the stalk. In Europe it is by far the most common form of celery, but has never
been extensively cultivated in the United States, though it is found in the larger
markets. The roots are white and more or less globular in shape, closely
resembling turnips in appearance. This vegetable deserves to be more widely
known, being extremely hardy and of easy cultivation. It is mostly used for
flavoring soups, except by the Germans who use it in the same manner as
potatoes for potato salad. Planted 7 or 8 inches apart and 3 feet between the
rows it will yield abundantly, and succeed best where celery will. The edible
portion develops into a bulbous root weighing 4 to 6 ounces when trimmed, and
these bulbs when properly packed away in the cellar will keep almost until
spring. Where the ground but slightly freezes, the plants may be safely left
unharvested for spring use.—(F. B. 255, 295; Mich. E. S. 20.)

CELERY.
The ideal climatic conditions for the production of celery are bright sunshine,
pure air, cool nights, and a well-distributed rainfall of about 8 inches during the
growing period in the field or garden.
Soils.—In the production of celery for domestic use, a rich, mellow, sandy
loam will give the best results. The soil of the seed bed should contain plenty of
leaf mold and should be passed through a sieve having not less than six meshes
to the inch. The soil of the transplanting bed need not be sifted so fine, and
some well-rotted barnyard manure should replace a part of the leaf mold; in
other respects it should be the same as that of the seed bed. Any fertile, well-
drained soil will grow celery, but a loose, sandy loam is preferable. If nothing but
clay soil is available, it may be made to produce good celery by the liberal
application of well-rotted barnyard manure. On clay soils there is likely to be
injury caused by the soil becoming washed into the hearts of the plants while
they are yet small.
Fertilizers.—For the production of the home supply of celery there is no
fertilizer that is so satisfactory as well-rotted barnyard manure. In many localities
the supply of manure is limited, and it may be necessary to depend almost
entirely upon commercial fertilizers. If fresh stable manure is used, it should be
plowed under in the autumn. If the manure is well rotted, it may be plowed
under early in the spring or used as a top-dressing a short time before planting
in order to bring the manure to the surface. From 10 to 20 tons of manure to the
acre should be applied each year that the land is planted to celery. The
application of lime will improve most soils. Following the use of stable manure an
application of 1,000 pounds of ground quicklime as a top dressing will be
beneficial. Soils that are liable to leach during the winter can be held by planting
to rye and the crop turned under quite early in the spring. When applied to clay
soils the lime has a tendency to lighten them, and sandy soils are rendered more
retentive of moisture by the addition of lime. An application of 500 to 800
pounds of common salt to the acre is considered desirable by some growers.
Celery will take up a limited quantity of salt, and its flavor is improved thereby.
One to 2 tons of high-grade fertilizer to the acre may be profitably applied on
most soils in addition to the stable manure and lime. As a rule, the quick-acting
fertilizers are used, and a mixture suitable for growing celery should contain
about 6 per cent of nitrogen, 5 per cent of available phosphoric acid, and 10 per
cent of potash.
Time and Method of Plowing.—As a rule the land should be plowed several
weeks before planting. At the North it is desirable to plow the celery land in the
autumn and allow the soil to lie exposed to the action of frost during the winter.
At the South it will be necessary to plow but a short time before planting. The
plowing should be very thorough, and in most cases with a somewhat heavier
plow than that generally used for other crops.
Smoothing and Pulverizing.—A few days before the land is required for
planting, the surface should be cut with a disk or cutting harrow, followed by
such tools as are necessary to pulverize the soil to a depth of 5 or 6 inches. Just
before planting, the land should either be rolled or gone over with a float, or
drag, made by nailing together planks or scantlings, in order to secure an even
surface for planting.
Marking Rows.—The rows in which the celery plants are to be set should not
be marked until a short time before planting, in order that the soil may remain
fresh. A marking device similar to the ordinary corn marker may be used, but
some form of roller with a number of projecting pegs to form holes in which to
set the plants is desirable. A device of this character can be constructed by
replacing the wheel of an ordinary wheelbarrow with a roller having a series of
pegs.
Selection of Seed.—The first and most important consideration when
preparing to grow a crop of celery is the securing of good seed, not merely seed
of which a large percentage will germinate, but that having strength and vigor
sufficient to give the seedling a good start. As the seeds of celery are very small,
it is necessary that only a small percentage of the number usually sown should
actually grow in order to secure an abundance of plants; but as low germination
and the necessary vigor are seldom both to be found in the same packet of seed
that seed which has a high percentage of germination is preferable.
Sowing for an Early Crop.—For sowing seed during the early part of the
season, the plan best suited to the requirements of the farmer or amateur
grower of celery is to secure a wooden flat or tray about 16 by 24 inches in size
and 3 inches deep, with several small holes in the bottom for drainage. After
filling with sifted soil level it off even with the top, and either shake down the soil
or press it down by means of a board before the seeds are sown. Either sow in
drills 2 inches apart or scatter broadcast, and cover the seed by sprinkling
through a fine sieve a very small quantity of leaf mold or sand. This tray can be
placed in the window of a moderately warm room in the dwelling, and the soil
should be watered by sprinkling very lightly as often as necessary to keep the
surface from showing dryness, but the soil should not become waterlogged.
Sowing for a Late Crop.—The method now in use by most large growers is to
prepare a tract of land by pulverizing with horse tools and then raking by hand,
after which the seed is sown broadcast by means of a wheelbarrow grass-seed
drill. The soil is sometimes pressed down with a plank after the seeds are
scattered, but some growers maintain that there is a decided advantage in
leaving the soil slightly uneven, as the seeds fall into the shaded places and are
protected from the direct rays of the sun. The seed will become sufficiently
covered by rains or by watering. Should more than 20 per cent of the seed
usually sown germinate, it is necessary to thin out to prevent overcrowding, with
its attendant injury. To prevent the surface of the soil becoming too dry, it may
be necessary to partially shade the young plants during the warm days of early
summer, but the shading should never be so dense as to cause them to become
"drawn."
Transplanting.—In case the grower adopts the plan of transplanting twice, the
seedlings will be ready for the first handling in from four to six weeks from the
time the seed is sown. The seedlings may be transplanted to trays or to beds in
the open ground. This transplanting answers two purposes: (1) The seedling
plant of celery has a straight root, or taproot, which is broken in transplanting,
causing a large mass of fibrous roots to be formed. In the case of a plant allowed
to remain in the seed bed until planting-out time this taproot has gone far down
into the soil and the plant has formed very few side roots; consequently it suffers
a great shock in the process of planting in the field, and a large number of plants
will need to be replaced. (2) When transplanting twice is practiced there is no
necessity for thinning, and a more uniform lot of plants is obtained. Two
handlings can not be recommended when celery is grown on a large scale, as the
cost of labor is too great. It is better to have a surplus of plants and to renew
those that fail.
Watering.—When the seed bed is prepared, the soil of which it is composed
should contain as much moisture as possible and yet be in good condition to
handle. After sowing and covering the seeds the bed should be sprinkled lightly.
During the period between seeding and the appearance of the plants the bed
should be watered only as often as it shows indications of dryness; however, the
surface should never become dry. During the first few days a moist cloth may be
spread over the surface of the seed bed in order to conserve the moisture, but
this covering should be removed before the seedlings begin to appear. After the
plants are up, care should be taken not to water too heavily, as the seedlings are
liable to "damp off"; but the ground should never become so dry as to check
their growth. Celery requires the most water while making its greatest growth,
which occurs late in the summer. As the crop approaches maturity the water
should be applied sparingly, and it should be withheld altogether for some time
before blanching. Among the methods of applying the water, the most simple
and usually the most desirable practice, especially where the surface of the soil is
even, is to run the water along the rows by means of small furrows, 8 or 10
inches distant on either side of the row. This method is well adapted to use on a
gentle slope with the rows running up and down the incline. When the water is
sprinkled over the entire surface it should be done late in the day, so that the soil
may, during the night, absorb the moisture and prevent a crust being formed, as
would be the case were the water applied under the direct heat of the sun.
Growing Without Irrigation.—For a home supply of celery it is often possible
to select a rather moist but well-drained piece of land whereon it may be grown
without artificial watering. In this case the plants should be set while the
atmosphere is filled with moisture, preferably between gentle showers, and the
moisture afterwards retained in the soil by frequent shallow cultivation or by the
application of a mulch around the plants. This method can not be followed in
climates where irrigation is necessary for the production of crops, but is
applicable in regions that have an ordinary rainfall during the growing season.
Planting.—For domestic use, where plenty of land is available, it will be found
most economical to plant in single or double rows 4, 5 or 6 feet apart, with the
plants 5 or 6 inches apart in the row. If the space is limited, solid beds about 5
feet wide will be found suitable, with the plants set 7 or 8 inches apart each way.
By planting in rows the crop may be worked with a horse cultivator or a wheel
hoe and the banking more easily done, and thus the cost of production is
lessened. With the solid-bed system the work must all be done by hand. If
possible, the planting should be done when the soil is rather moist and the
atmospheric conditions suitable to the subsistence of the plants until the roots
can again furnish sufficient moisture to supply them. The bed should be
thoroughly watered a few hours before the plants are removed, and a knife or
trowel should be run between the plants so that they may be lifted with a clump
of earth and with most of their roots attached.
Mulching.—In muck soils it will not be found necessary to mulch the ground
around the plants after setting, but some kind of a covering is desirable on sandy
and clay soils. As soon as the plants are in position and before any water is
applied, cover the ground for a distance of 8 or 10 inches on either side with any
finely divided material that will shade the top of the soil and prevent a crust
being formed after watering; half-rotted manure is preferable for this, as it aids
the growth by its fertilizing qualities. Good celery can be grown on clay upland
with but one watering—at the time of planting—provided that plenty of mulch is
applied as soon as the plants are set. The roots of celery, after it is once
transplanted, run close to the surface, and the mulch will protect them from the
heat of the sun. Among materials that may be used for a mulch may be
mentioned pine needles, leaves of any kind, straw, cornstalks run through the
cutter, clippings from the lawn, etc., none of which, however, are as good as
barnyard manure. Have the material to be used as a mulch near at hand, and as
the plants are set cover the soil around them to a depth of 2 inches, bringing the
mulching material up close to the plant, but being careful to allow none to get
into the heart. Apply the mulch before watering, if possible.
Where celery is planted in single rows and mulched it will only be necessary
to maintain shallow cultivation between the rows, not allowing the cultivator
teeth to come nearer the plants than the edge of the mulch. Where no mulch is
used the cultivation may be carried a little closer to the plants, but should be
very shallow, and at no time should deep cultivation be practiced, as the roots
are to be found very near the surface of the soil. If a mulch is used no hand
cultivation will be required, either along the side or between the plants in the
row, except to pull any weeds that may spring up. Where no mulch is used it will
be necessary lightly to stir the surface with a wheel hoe or iron rake, to prevent a
crust being formed after each rain or watering. Keep the surface of the soil
smooth and in no case allow lumps of earth to remain near the plants.
Blanching.—In its original wild state the stems of celery are tough, full of
woody strands, of a rank flavor, and green in color, being similar to the outside
stems or trimmings of our present varieties. The object of blanching is to secure
leafstalks free from woody strands, crisp and tender, and without the rank flavor
found in those that are green. Of the cultivated plant there are two classes of
varieties, the large-growing, or giant, and the dwarf sorts. These are again
divided into those which must be blanched by excluding all the light and those
which are in a measure self-blanching. Of the former the Giant Pascal variety is a
type, and of the latter the Golden Self-Blanching variety is a good illustration.
Blanching is accomplished by the same general method that is employed for
destroying the coloring matter in any plant tissue, that is, by excluding the light
and allowing the growth to proceed in the dark. The particular method to be
adopted must be determined largely by the time when the crop is to be used. If
for early use or marketing, the blanching must be completed where the plants
are grown; but if the celery be for winter use the blanching may take place after
the crop has been removed from the field and placed in storage. In fact, it is best
to blanch as little as possible before storing when the product is to be kept until
late, as the keeping qualities are better while it is unblanched. When planting for
early use it is necessary to choose one of the self-blanching varieties, such as
may be conveniently blanched by the use of boards or other similar means.
For early blanching on a small scale, such as would be employed on the farm
or in the garden of the amateur horticulturist, there are several methods. One of
the most common is by means of boards placed on edge along each side of the
row.
After the boards are in position it is a good plan to run a celery hiller between
the rows and to throw a little soil against the lower edges of the boards to close
any openings that may result from the uneven surface of the soil.
Two or three weeks' time will be required to complete the blanching of the
early varieties, and the boards must be kept in position until the crop is removed
from the ground, after which they may be used again two or three times during
the season. If the celery is allowed to remain in the boards too long after it has
reached a marketable stage, it loses in weight and flavor and is liable to be
injured or even destroyed by the attacks of blight. This is especially true during
the earlier part of the season, when the weather is warm. At the end of the
season the boards should be piled flat, with strips inserted at every fourth or fifth
course, and the whole pile roofed over to shed off rain; treated in this manner
they will last from ten to twelve years.
Perhaps the most satisfactory way of blanching early celery on a small scale is
by means of ordinary farm drain tiles of about 4 inches inside diameter, placed
over the plants after they have become almost fully grown. To facilitate the work
of placing the tiles over the plants, some of the outside leaves should be pulled
away and the main part of the plant loosely tied together by means of a soft
string, or, better, with what is known as paper twine, being a string made by
twisting a strip of soft paper. This string will lose its strength as soon as it
becomes wet, and will offer no resistance to the further growth of the plant. If
the common, unglazed tiles are used the evaporation from their surface has a
tendency to keep the plant cool during the heat of the day, and a very crisp and
tender product is the result. This method of blanching is desirable also on
account of its cleanliness, as celery treated in this way will need very little
washing before marketing.
The most common method for blanching celery on a small scale is that of
banking with soil, and it is by this means that the finest flavor can be obtained.
Where the plants are set in single rows the soil can often be partially thrown up
by means of a plow, or, better, by a celery hiller. Before the plow or banking
machine is used a small quantity of dirt must be placed around the plants by
hand to hold them in position while the earth is being thrown around them. This
may also be accomplished by tying up the plants with paper twine, as previously
recommended for use in connection with tiles.
Storing.—The plan usually adopted where but a small quantity of celery is to
be stored for winter use is to bank up with earth and cover the plants where
grown. Place enough earth around the base of the plants to hold them in good
form, and then allow them to remain without any further banking as long as
there is not danger of a hard frost. Celery may be safely stored in cellars
provided the temperature is kept low and plenty of ventilation maintained. The
warmth and dampness of the ordinary cellar have a tendency to cause the celery
to decay, but these conditions can frequently be overcome. Celery will readily
absorb any odor that may be present in the atmosphere of the storage place,
and care should be taken to provide sanitary conditions. When storing in a cellar,
the plants should have most of their roots attached, and a bed of moist sand in
which to set them should be provided.

CELERY BANKED WITH EARTH TO BLANCH IT

Preparing Celery for Market.—In preparing it from the rows where grown, it is
not necessary to remove the entire root from the earth, but it may be cut off just
below the surface of the soil by means of a stiff knife. Remove the outside leaves
and trim the root evenly, pack in boxes, and load on the wagon for removal to
the washing house. The blanching boards should not be removed till necessary,
and the trimmed celery must not be allowed to lie exposed to the sun or wind for
any length of time. It is well also to have a piece of canvas to protect the celery
while it is on the wagon on the way to the washing house. In marketing from the
trenches the process is practically the same as from the rows, except that the
celery is already loosened from the soil and the roots can be removed more
easily. Upon reaching the washing room the celery is placed upon a rack
consisting of wooden slats over a large trough and subjected to a spray of cold
water to cool it and to remove the adhering soil. After washing, it is allowed to
drain; then it is tied in bunches of 12 or more plants each, according to the size.
The bunches are packed 6 in a box for first-grade celery and 8 or 9 for second or
third grades. These boxes should be practically air-tight, and a lining of paper
should be placed in them before packing the celery, or each bunch should be
wrapped separately. The celery should be nearly dry before it is placed in the
boxes, and throughout the entire handling must be kept as cool as possible.
Sanitary Conditions.—It is essential that the celery should be washed in pure
water to prevent the transmission of disease germs. Any germ, such as that
producing typhoid fever, which is found in contaminated water, is readily carried
to the digestive system of the consumer, and may or may not produce an attack
of the disease, according to the strength of the person to resist it. The
washhouse and its surroundings should be kept clean and free from any
decomposing materials. Shippers and dealers alike lose sight of the fact that the
edible portion of celery is constantly being exposed to the contaminating effects
of dirty wagons, unclean cars, and dusty markets. Many persons have
discontinued the use of celery on account of the unclean condition in which it is
served. This statement holds good for all vegetables that are served in the raw
state, but it is especially applicable to celery.
Estimates of Returns.—Anyone contemplating making a start in celery
growing will do well to first investigate the market prospects, and unless
satisfactory shipping arrangements can be made beforehand the crop should be
planted only on a small scale for one or two years, until a local trade can be
established. It is fair to estimate a return of 1,500 dozen from 1 acre; and this
should bring 25 cents per dozen, at the lowest average estimate; this will yield a
gross income of $375 to the acre, leaving a net balance of $125 to cover the
interest on the investment and the profit. As a matter of fact, the growers who
are making a success of celery raising—and many are doing so—receive a net
profit of $100 an acre over and above the interest on the investment. On the
other hand, hundreds of acres are grown annually which do not much more than
pay expenses, but this is due to the fact that the soil has become exhausted and
the product is consequently undersized and inferior.—(F. B. 255, 282; Cornell E.
S. 132; Colo. E. S. 144.)

CETEWAYO, OR ZULU, POTATOES.

 The Cetewayo, or Zulu, potato, a wild variety of Solanum tuberosum found in
Africa, is sometimes grown as a garden vegetable for its flavor and novelty. It
has practically the same percentage composition as the ordinary potato. When
cooked, the flesh is purple in color, but when brought in contact with vinegar, as
in salads, it turns red.—(F. B. 295.)

THE CHAYOTE.
The chayote suggests the cucumber rather than any other of the cultivated
plants of the same family, but is a larger and more vigorous plant, climbing
widely by means of numerous branched tendrils. When grown under ordinary
garden conditions the cultural requirements of the chayote may be said to be two
in number: (1) A somewhat sheltered situation and (2) something to climb upon.
While the vine will not refuse to grow without these advantages, the results will
not be satisfactory. Like many climbing plants, the chayote is very susceptible to
injury from the wind, while, unlike many Cucurbitaceae, it does not seem to take
kindly to creeping upon the ground, at least in the Tropics. In the different parts
of the world the chayote has been found to grow upon a great variety of soils,
though it is generally considered to thrive best in a loose sandy or loamy
substratum, providing sufficient humus or other fertilizing material be at hand.
Although it has been found possible to secure plants from the seed when planted
alone, or even from the embryo when carefully extracted from its seed coats, it is
the universal practice to plant the entire fruit. The fruit should be gathered
before fully matured, because of the tendency to germinate. It is like the
cucumber, edible at any stage of growth, and may be picked when large enough.
The chayote is a good shipper and may be shipped in bulk in vegetable crates,
wrapped and well packed; cold storage will not be necessary.—(Dept. Ag., Div. of
Botany 28; P. Rico A. E. S. 7).

CHERVIL.
Under the name of chervil two distinct plants, known as salad chervil and the
turnip-rooted chervil, are cultivated. The seeds of the salad chervil are sown in
spring and the crop will thrive on any good garden soil. The seeds of the turnip-
rooted chervil should be sown in the early autumn, but they will not germinate
until the following spring. The edible part of this plant is the root, which
somewhat resembles the carrot and is used in the same manner. The leaves are
used the same as parsley for garnishing and in flavoring soups.—(F. B. 255.)

CHICORY.
 Chicory is grown for two or three purposes. The root of this plant is the
common adulterant of coffee, and large quantities are used for this purpose. The
commercial growing of chicory is confined to a few sections, as the crop will not
thrive on every kind of soil. A deep, rich loam, without excessive amounts of clay
or sand, is desirable, and soil that is not too rich in nitrogenous matter is best
suited to the production of roots. The roots are frequently placed in soil under a
greenhouse bench or in a warm cellar and covered with a foot or more of straw,
or with a light covering of straw and then several inches of warm manure. Under
this covering the leaves will be formed in a solid head, which is known on the
market as witloof. Chicory has run wild in some parts of the country and is
considered a bad weed. The handsome blue flowers, which are borne the second
season, are very attractive. As a pot herb chicory is used like spinach, but the
leaves should be boiled in two waters to remove the bitter taste. As a salad the
roots are dug in the autumn and planted in cellars or under a greenhouse bench,
where they produce an abundance of blanched leaves, which are eaten raw. The
blanched leaves are also boiled and used as greens.—(F.B. 255; U. Idaho E. S.
10.)

CHILE.
The chile is used in many different ways and it is quite an important article of
food among the Spanish speaking population in the Southwest and in Mexico. It
is eaten both in the green and ripe state. It may be grown on ridges or in level
plats. The former method is the more common in New Mexico. In the spring after
the ground has been plowed and leveled (the plowing of the land can be done in
the fall or winter) and just a little before planting the ridges are made. These
ridges may vary in height from 8 to 12 inches. It is better to irrigate the ridges
before planting, though this is not always done. The object of irrigating before
planting is to get the water mark on the side of the ridges and to settle the newly
plowed soil somewhat. As soon as the soil is dry enough so it can be worked,
which is generally from four to seven days, the seed is planted usually on one
side of the ridge and just above the water mark. The seed is planted by hand in
hills about every two feet in the row. The chile does not stand freezing weather,
though it will stand a little more cold than tomatoes. For the convenience of
intending chile growers the following table which gives the number of hills per
acre at different distances has been prepared:

Number of
Distance. Hills per Acre.
3½ feet between rows × 2 feet in the row 6222
3½ feet between rows × 2½feet in the row 4978
 4 feet between rows × 2 feet in the row 5445
4 feet between rows × 2½feet in the row 4356
Planting.—The seed is planted on the side of the ridge, when the ridge
method is practiced. The southern exposure of the ridge is always preferable
since this is usually warmer and the germination, other factors being uniform, is
quicker. If level culture is practiced there is no choice of exposure. Whatever
method of planting is followed care should be taken not to bury the seed too
deeply. As a general thing the seed should not be deeper than three-fourths of
an inch to an inch and a half. Shallower planting, if the moisture is kept normal,
will give quicker and better germination. More seed is required per acre when the
planting is done by hand on the ridges than when it is drilled with a garden drill
in plats.
Thinning.—Chile started from seed planted in the field must be thinned to one
or three plants to the hill. When the chile has been thinned out properly the plant
or plants in the hill branch out considerably and produce a heavier and better
crop. If too many plants are left to the hill there is a marked tendency for the
plants to grow too tall and more or less top heavy. The chile is thinned out when
about 3 to 5 inches high. If a good germination takes place it is more difficult to
thin the chile, because there are more small plants to the hill to be pulled out.
Care should be had in selecting the strongest plants in the hill and in injuring as
little as possible the roots of those which remain. While the common way of
growing chile is to plant the seed out in the field in the spring, it can also be
grown by starting the plants in cold frames early in the season and transplanting
to the field as soon as danger of frost is over.
Irrigation.—After the irrigations to get the crop started have been given, the
frequency of the subsequent irrigations depends upon the weather and soil
conditions, and for that reason no specific statement can be made just when and
how often the chile should be irrigated. One thing, however, is important to keep
in mind, and that is that the chile plant keeps bearing as long as it is growing. If
the growth should be checked by the lack of irrigation the plant stops bearing
and the blossoms and the very small pods are likely to drop off. The grower
himself should study his local conditions and decide for himself when and how
much to irrigate. While the chile plant resists considerable drought, at the same
time, it should not be allowed to suffer from the lack of irrigation. When the chile
is grown on ridges the space between the ridges should be allowed to fill with
water almost up to the plant. If the water is simply turned in and allowed to rush
down the furrow to the other end the ridges will remain practically dry,
necessitating frequent irrigations to keep the plants from suffering. In irrigating
the chile on ridges always aim to hold the water long enough in the furrow for
the ridges to get fairly well soaked through. In the level plat the irrigation is
more simple and the soil around each hill gets wet sufficiently while the water is
running down to the end of the plat. When the plats are quite long and are made
up of a series of squares as soon as each square is filled with water the border,
dividing that square from the next one, is cut and the water rushes into the next
square which is treated the same as the one before.—(N. Mex. Col. Ag. and
Mech. Arts 67.)

CHIVE.
This is a small onion-like plant having flat, hollow leaves which are used for
flavoring soups. The chive rarely forms seeds, and it is propagated by the bulbs,
which grow in clusters. The leaves may be cut freely and are soon replaced by
others.—(F. B. 255; S. Dak. E. S. 68.)

CITRON.
The citron is a type of watermelon with solid flesh which is used for preserves
and sweet pickles. The rind of the watermelon is frequently substituted for citron.
The cultivation of the citron is the same as for the watermelon.—(F. B. 255; U.
Idaho E. S. 10.)

COLLARDS.
The culture and uses of collards are the same as for cabbage and kale.
Collards withstand the heat better than either cabbage or kale, and a type known
as Georgia collards is highly esteemed in the Southern States. Collards do not
form a true head, but instead a loose rosette of leaves, which, when blanched,
are very tender and of delicate flavor.—(F. B. 255; U. Idaho E. S. 10; P. Rico A. E.
S. 7.)

CORN SALAD.
Corn salad is also known as lamb's-lettuce and fetticus. Sow the seed during
the early spring in drills 14 to 18 inches apart and cultivate the same as for
lettuce or mustard. For an extra early crop the seed may be planted during the
autumn and the plants covered lightly during the winter. In the Southern States
the covering will not be necessary and the plants will be ready for use during
February and March. The leaves are frequently used in their natural green state,
but they may be blanched by covering the rows with anything that will exclude
the light. Corn salad is used as a salad in place of lettuce, or mixed with lettuce
or water cress. The flavor of corn salad is very mild, and it is improved by mixing
with some other salad plant for use. It is also boiled with mustard for greens.—
(F. B. 255.)

CRESS.
Under the name of cress there are two forms, the water cress and the upland
cress. The upland cress, sometimes called peppergrass, is easily grown from
seed sown in drills a foot apart. As the plants last but a short time, it will be
necessary to make a sowing every few days if a continuous supply is desired.
Water cress can be grown all the year in small open ditches containing
running spring water. It is best and most easily produced in water from rather
warm springs in limestone regions. A sufficient supply for family use can be
grown in a small spring-fed brook, and the plants may be started either from
small pieces of plants or from seed. Cress is used in salads, to which it imparts a
pleasant pungency.—(F. B. 255; U. Idaho E. S. 10; P. Rico A. E. S. 7.)

CUCUMBERS.
Soil.—The soil best adapted to the cultivation of cucumbers in the open is a
light sandy loam, one which responds quickly to temperature and fertilizer. Such
soils are prepared early in the season and thrown into gentle undulations, so as
to produce slight ridges upon which to plant the seed to insure good surface
drainage.
Fertilizers.—The soil for cucumbers should be made very rich by the annual
application of heavy dressings of stable manure to be incorporated with the soil.
During the time it is not occupied by cucumbers or lettuce, cowpeas are
frequently grown upon the area and turned under prior to planting a fall crop of
lettuce. In addition to this, liberal applications of a fertilizer carrying a
considerable percentage of nitrogen are employed.
Planting.—There are almost as many methods of planting cucumbers as there
are growers. Some plant in hills the standard distance of 6 feet apart each way;
others plant in hills 6 feet apart in one direction and 2 or 3 feet apart in the row,
while others plant in drills or broad belts 6 feet apart and chop out the plants to
stand about a foot apart in the row after all danger from insect depredation has
ceased. The methods which seem most economical under the conditions at hand
will of course be adopted by the grower. In outdoor culture the cucumber is
frequently used as a companion crop to other crops, like beans. Beans being of
rapid growth come on quickly and form a partial protection or wind-break for the
young cucumber plants. When arranged in this way, cucumbers are planted in
drills or in hills 6 feet apart and a row of beans is placed between two rows of
cucumbers, a method which insures a very complete and satisfactory use of the
ground. The quick maturity of the beans allows them to be harvested and
entirely removed from the area before it is required for the cucumbers.
Harvesting.—Cucumbers intended for pickling purposes are harvested when
they have attained a length of from 2½ to 5 inches. Because such cucumbers
are bought by weight it will readily be seen that the small-sized pickles are less
profitable to the grower than the larger ones, and in order to secure them before
they have attained an unsalable size it is necessary that the picking be repeated
at frequent intervals, as cucumbers grow rapidly and a delay of twenty-four to
forty-eight hours in harvesting would render many of them unsalable. It is
therefore necessary to have regular intervals to harvest certain areas of the
patch and to continue this routine throughout the bearing season. Another point
which is of prime importance in the management of the cucumber patch is that
none of the fruits be allowed to come to maturity. The ripening process, which
means the development and maturing of the seeds, produces a heavy strain
upon the growing plant, the life and yield of the plant being in proportion to the
number of fruits which are allowed to ripen. If no fruits are allowed to come to
maturity the plants will remain green and in an active vegetative condition longer
and will produce a much larger aggregate number of fruits.
Dill Pickles.—Dill pickles, which are much prized and command the highest
price among pickles, can be made from fresh cucumbers as they come from the
vines, or from vat stock which has been carried for some time at the salting
station.
Cucumbers Grown in Cold Frames for Market.—Soil for use in cold frames
should be a well-enriched sandy loam of the type of the usual sandy loam. If it
can be dark in color, this is an advantage. If normally light, the color can be
changed by the addition of muck or by incorporating well-decomposed stable
manure with the surface soil. A dark color is of some advantage in helping to
raise the temperature in the frames under the glass.
Watering.—Since the glazed sash prevent the soil beneath them being
moistened by natural means—that is, by rain or dew—it is necessary that means
be provided for watering or irrigating the plants. This can be done by arranging
pipes upon the surface of the ground or at a convenient height overhead, so as
not to interfere with cultivation, from which water can be drawn to sprinkle the
surface of the beds at desired intervals and as the plants may require. The work
of watering should, however, be very carefully done. The same general
precautions necessary for the care of plants in cold frames should be observed—
that is, to do the watering in the morning on bright days only, when air can be
admitted and when the sun will soon dry the moisture from the leaves of the
plants. In this way much can be done to protect the plants from injury from such
diseases as the damping-off fungus and mildew.

JAPANESE CLIMBING CUCUMBER NEARLY SIX FEET FROM THE

GROUND

WELL-GROWN CUCUMBERS
 Ventilation.—Besides the precautions to be observed in watering plants in
cold frames, extreme care is necessary to give the plants sufficient air to keep
them in a healthy condition. If the atmosphere is allowed to become close and
very hot, the plants will be weakened and thus rendered more susceptible to the
attacks of plant diseases.
Forcing Cucumbers Under Glass.—Forcing is a technical term used by
gardeners to designate the growing of plants out of their normal season under
an artificial environment. The cucumber is one of the few garden plants which
lend themselves to this manner of cultivation in addition to their more extensive
cultivation in the open ground. Under the stimulus of forcing work, two distinct
types of cucumbers have been developed. These are recognized in the trade as
the English type and the American type. The English type is purely a product of
forcing-house conditions, as the climate of England is not congenial to the
growth and development of the cucumber in the open. The American type of
cucumber is primarily a product of field conditions, and the few varieties which
have been developed to meet the requirements of the forcing house are simply
modifications of the existing field or outdoor forms. The English type of
cucumber is a long, cylindrical, uniformly green fruit, with few seeds and a very
fleshy seed cavity; in fact, the normal seed cavity of the forced cucumber is
almost entirely wanting. The triangular shape characteristic of the normal
outdoor cucumber has been lost, and the cylindrical outline almost perfected.
There is considerable difference in the size and length of the various English
varieties of cucumbers. The American type of cucumber is primarily grown in the
field, the product to be used either for pickling or for slicing. Forcing cucumbers
in America is confined to those varieties which produce large fruits suitable for
slicing. Only three or four of the better and larger field varieties are adapted to
this purpose. Notable among these is the White Spine, the Arlington White Spine
being the variety which has been especially developed for forcing. The Long
Green, or a modification of it, is also sometimes used, but aside from these two
varieties there are few that ever find their way into the forcing house. Such
varieties as the Boston Pickling, Chicago Pickling, and the cluster varieties in
general are not adapted to forcing purposes. The forcing of cucumbers
presupposes that an adequate forcing house or greenhouse is at hand for such
work. The chief desideratum in a forcing house for cucumbers is a maximum
amount of light, sufficient headroom, and adequate radiation to maintain a
temperature varying from 65° to 85° F. The amount of radiation will, of course,
depend upon the style of heating employed, whether steam or hot water, and
upon the location of the greenhouse, whether at the north or the south; the
outside temperature determining to a considerable extent the amount of
radiation required in the house to maintain a given degree of heat.
 Propagation.—There are a number of methods of propagation followed by
successful cucumber growers, all of which have some advantages. Three of the
more common practices are as follows: (1) To plant the seeds of cucumbers in
the soil of the bench where the plants are to grow and mature; (2) to plant the
seeds of the cucumbers in 3-inch or 4-inch pots filled about half full of soil and
after the seeds have germinated and the hypocotyl or stem of the seedling has
elongated to fill the pots well up to the seed leaves with soil; and (3) to plant the
seeds in cups similar to those used for harvesting strawberries, except that the
cups for this purpose are usually made of Georgia pine. In the first case, where
the seeds are planted directly in the soil on the benches, cucumbers are usually
employed as a crop to follow lettuce, seeds being planted in the lettuce benches
before the crop is entirely removed, heads of lettuce being taken out at proper
distances to allow for the correct spacing of the cucumber plants, and the seeds
of cucumbers planted in the areas so left. In the other two cases the rearing of
the plants for forcing purposes can be carried on in a small house especially
designed for this purpose or in a general propagating house, thus obviating the
necessity of heating and maintaining normal conditions in the growing house
during the period previous to which the plants begin to run.
Planting on the Benches.—As soon as the plants show well-developed runners
and are 10 to 12 inches long they should be placed in their permanent position
upon the greenhouse benches. Plants grown in pots must be carefully removed
from these receptacles to the bench, but those grown in the wooden cups above
referred to can be planted, cup and all, in the soil of the bench. The utmost care
should be exercised to keep the plants of the cucumber growing rapidly at all
times. If cucumbers receive a severe check or are placed under conditions which
are not entirely congenial to them, they are liable to become dwarfed and
stunted, and as soon as vigorous growth ceases they become the prey of the
melon aphis, mildew, and other pests and diseases which are so annoying to
growers of cucumbers under artificial conditions.
Distance to Plant.—After the plants have attained a height of 10 or 12 inches
and are in a vigorous growing condition they should be placed about 15 or 18
inches apart in single rows upon the side benches of the greenhouse, which are
normally 3½ feet wide, or if planted on 8-foot benches they should be planted
about 10 or 12 inches from the edge of the bench and 15 to 18 inches apart and
parallel with the edge of the bench. In the broad benches, where more than a
double row can be carried, plants can be set about 18 inches apart and in rows
about 2 feet apart. A satisfactory plan for an 8-foot bench will be a row parallel
with and 10 inches from each edge of the bench and a double row 18 inches
apart through the middle of the bench. It is well, however, to allow as much
space as possible. The cucumber is a rank-growing plant and many side branches
will develop if sufficient space is allowed.
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

textbookfull.com