Get ebook downloads in full at ebookmeta.

com

Graph Neural Networks in Action (MEAP Version 4)

Keita Broadwater

https://ebookmeta.com/product/graph-neural-networks-in-
action-meap-version-4-keita-broadwater/

OR CLICK BUTTON

DOWNLOAD NOW

Explore and download more ebook at https://ebookmeta.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Graph Neural Networks: Foundations, Frontiers, and

Applications First Edition Lingfei Wu

https://ebookmeta.com/product/graph-neural-networks-foundations-
frontiers-and-applications-first-edition-lingfei-wu/

ebookmeta.com

Google Anthos in Action - MEAP Version 6 Antonio Gulli

https://ebookmeta.com/product/google-anthos-in-action-meap-
version-6-antonio-gulli/

ebookmeta.com

Statistical Analysis of Graph Structures in Random

Variable Networks V. A. Kalyagin

https://ebookmeta.com/product/statistical-analysis-of-graph-
structures-in-random-variable-networks-v-a-kalyagin/

ebookmeta.com

These Fleeting Shadows 1st Edition Kate Alice Marshall

https://ebookmeta.com/product/these-fleeting-shadows-1st-edition-kate-
alice-marshall/

ebookmeta.com
Lights Camera Murder Reality TV Cozy Mysteries 1 1st
Edition Nikki Haverstock

https://ebookmeta.com/product/lights-camera-murder-reality-tv-cozy-
mysteries-1-1st-edition-nikki-haverstock/

ebookmeta.com

Electronic Waste Management: Policies, Processes,

Technologies, and Impact 1st Edition Sunil Kumar

https://ebookmeta.com/product/electronic-waste-management-policies-
processes-technologies-and-impact-1st-edition-sunil-kumar/

ebookmeta.com

Weird Fiction: A Genre Study Michael Cisco

https://ebookmeta.com/product/weird-fiction-a-genre-study-michael-
cisco/

ebookmeta.com

Tales from Hans Andersen - Pearson English Readers Level 2

2nd Edition Hans Andersen

https://ebookmeta.com/product/tales-from-hans-andersen-pearson-
english-readers-level-2-2nd-edition-hans-andersen/

ebookmeta.com

Sacred Sites and Sacred Stories Across Cultures:

Transmission of Oral Tradition, Myth, and Religiosity
David W. Kim
https://ebookmeta.com/product/sacred-sites-and-sacred-stories-across-
cultures-transmission-of-oral-tradition-myth-and-religiosity-david-w-
kim/
ebookmeta.com
Conceptual Review of Radiology 3rd Edition Mayur Arun
Kulkarni

https://ebookmeta.com/product/conceptual-review-of-radiology-3rd-
edition-mayur-arun-kulkarni/

ebookmeta.com
Graph Neural Networks in Action
1. Copyright_2020_Manning_Publications
2. welcome
3. 1_Discovering_Graph_Neural_Networks
4. 2_System_Design_and_Data_Pipelining
5. 3_Graph_Embeddings
6. 4_Graph_Convolutional_Networks_and_GraphSage
7. Appendix_A._Discovering_Graphs
MEAP Edition

Manning Early Access Program

Graph Neural Networks in Action

Version 4

Copyright 2020 Manning Publications ©Manning Publications Co. To

comment go to liveBook https://livebook.manning.com/book/graph-neural-
networks-in-action/discussion

For more information on this and other Manning titles go to

manning.com
welcome
Hello,

Thanks for purchasing Graph Neural Networks in Action. If you want a

resource to understand and implement GNNs, this book is for you! When I
was dipping my toes into GNNs, most resources were tutorials on
Medium.com, GNN library documentation, some papers with code, and a few
scattered videos on youtube. These were useful sources of knowledge, but
learning from them was inefficient.

These resources were scattered about the internet. And, in many of them,
there were unspoken assumptions and prerequisites that presented roadblocks
to comprehensive understanding. Finally, many tutorials and explanations
lacked the nuance to allow me to answer the questions of discernment like,
‘why use a GCN over GraphSage ?’.

I wrote GNNs in Action to remedy many of these pain points. I structured it

to be both a book where you can jump into a topic of interest and get
something down in code, and as a reference that comprehensively covers the
relevant knowledge that would allow one to employ and discuss GNNs with
confidence. This is a hard balance to strike, and I hope to have done it well.

Speaking of implementation, one other thing that bothers me is when a book

uses frameworks and languages that I don’t use. Though I use python
exclusively as the programming language, examples can be found that are
grounded in the two major frameworks, pytorch and tensorflow. For smaller
datasets, I use Pytorch Geometric and DGL; for scaled problems, I use
Alibaba’s Graphscope, as well.

If you are an intermediate user of python, and havesome data science or

machine learning engineering experience, you should be good enough to dive
in. I also assume you have basic knowledge of neural networks and linear
algebra.
I hope this book is a valuable resource for you to get your work project into
production, to have informed discussions with colleagues, or to make your
own toy projects and tutorials. Your frank feedback, comments, questions,
and reviews are welcome in the livebook discussion forum, so that this work
can be improved.

Happy Reading,

Keita Broadwater

In this book

Copyright 2020 Manning Publications welcome brief contents 1 Discovering

Graph Neural Networks 2 System Design and Data Pipelining 3 Graph
Embeddings 4 Graph Convolutional Networks and GraphSage
Appendix A. Discovering Graphs
1 Discovering Graph Neural
Networks
This chapter covers
Defining Graphs and Graph Neural Networks (GNNs)
Understanding why people are excited about GNNs
Recognizing GNN Use Cases
A big picture look at solving a problem with a GNN

“Sire, there is no royal road to geometry.”

Euclid

For data practitioners, the fields of machine learning and data science initially
excite us because of the potential to draw non-intuitive and useful insights
from data. In particular, the insights from machine learning and deep learning
promise to enhance our understanding of the world. For the working
engineer, these tools promise to deliver business value in unprecedented
ways.

Experience detracts from this ideal. Real data is messy, dirty, biased.
Statistical methods and learning systems have limitations. An essential part of
the practitioner’s job involves understanding these limitations, and bridging
the gap between the ideal and reality to obtain the best solution to a problem.

For a certain class of data, graphs (also called networks, which we will use
interchangeably in this book), the gap has proven difficult to bridge until
recently. Graph data has characteristics that have presented problems for
traditional machine learning and deep learning methods to penetrate. Yet,
graphs are a data type that is rich with information. They are also ubiquitous:
We can find network structures in nature (molecules), society (social
networks), technology (the internet), and mundane settings (roadmaps). In
order to use this rich and ubiquitous data type for machine learning, we need
a specialized form of neural network dedicated to work on graph data: the
graph neural network or GNN.

Some basic definitions and an example will help us get a better idea of what
GNNs are, what they can do, and where they can be helpful.

1.1 What are Graphs & Graph Neural Networks?

First, let’s define graphs and graph neural networks more precisely.

Graphs are data structures whose elements are relationships (expressed as

edges) between entities (expressed as nodes, also called vertices). In graphs,
the links and relationships in our data are first class citizens, and are key to
how we model the world, analyze data, and learn from it. This is in contrast
to the grid-like data that is common in traditional learning and analytics
problems, like the relational database table, pandas dataframe, or
excel/google sheet. The relationships in such grid-like data are much more
muted than in networks or graphs.

Figure 1.1 A graph. Circles are nodes, also called vertices, and lines are labels, also known as
edges. The numbers denote node IDs.

A graph neural network (GNN) is an algorithm that allows you to represent

and learn from graphs, including their constituent nodes, edges and features.
Graph neural networks provide analogous advantages to conventional neural
networks, except the learning happens on graph data. Applying traditional
machine learning and deep learning methods to graph data structures has
proven to be challenging. Graph data, when represented in grid-like formats
and data structures, has qualities that prevent consistent application of
machine learning methods that don’t take this into account (an example is
called permutation invariance, defined as the ability of a machine learning
method to be influenced by the ordering of the input graph representation). In
addition, traditional methods have failed to account for the network structure
in the learning process. GNNs can address both of these issues, which will be
explored in more depth in this book.

The concept of applying specialized neural network architectures to graphs

was first introduced in the late nineties, gained traction in the early 2000s,
and greatly accelerated after 2015 ([Wu] has a good list of these papers).
Today, there are a few examples of enterprise systems that employ GNNs at
scale. This is still a rapidly developing field, with many challenges remaining

Before we introduce graphs and GNNs further, let’s cast a classic machine
learning problem in a new light by using graphs. By doing so, we can get an
idea of how such problems can be represented as networks, and see possible
applications of GNNs.

For a brief example of a dataset that has historically been cast in a grid
manner, but is full of unexplored links and relationships, we can examine the
Titanic dataset, whose observations span the passengers of the doomed ship,
and label their survivorship. Most machine learning challenges and projects
based on this dataset express it in terms of the target variable (survivorship)
and columns of features, which fit well in a table or dataframe (figure 1.2).
Each row of such tables represents one person.

Figure 1.2. The Titanic Dataset is usually displayed and analyzed using a table format.

As expressive as tables like this are, their account of links and relationships
are superficial at best. In particular, they fail to convey the social links
between the people, the corridors that linked locations on the ship, and the
communication network.

First, the people on the ship shared multiple types of relationships, including
marital and blood relations (married, engaged, parent/child, siblings, and
cousins), business and employment relationships, and, of course, friendships.
Many table versions of the Titanic dataset give boolean indicators or counts
of immediate family relationships (e.g., in the table above, the alone feature is
True if a person had immediate family on the ship).

Network representations of social relationships can add depth via specificity.

For example, having relatives on the ship may be a factor in favor of survival,
but having socially important relatives would probably give a greater chance
at survival. Social networks can convey social importance via their structure.
An example of a graph representation of families on the Titanic is shown in
figure 1.3.

The next example of a network on the Titanic is the ship’s corridors,

consisting of connected hallways, stairways, junctions and landings, and the
adjoining cabins and workrooms can also be represented as a graph. These
are important to the survival question because the ease to quickly get to a
lifeboat in a crisis depends on one’s starting location. From certain cabins and
decks, it would be easier to reach a lifeboat in time than others.

The third example of a network on the Titanic is the ship’s communication

network, whereby the ship’s crew communicated between themselves, to the
passengers, and to the outside world (figure 1. 3). This is relevant to the
question of survival, because critical information about the crisis would reach
individuals only in proximity to communication nodes. The communication
networks of the Titanic and surrounding ships (both by wireless telegraph and
by analog signals, such as flares), also impacted survivability. There were
two ships within 14 miles of the Titanic when it sank. They saw the distress
flare, but for different reasons, decided not to help. If they were linked by
wireless, the compelling information about the rapid sinking of the ship may
have swayed one or both ships to attempt a rescue.

Figure 1.3. The Titanic Dataset. Left: with family relationships visualized as graphs (credit: Matt
Hagy). Right: An old illustration of the wireless communication between the Titanic and other
ships. It’s easy to interpret the ships as nodes and the wireless connections as edges.

Assuming one had access to this graph data and it was of sufficient quantity
and quality, what could a GNN glean from it? First, since the historical
records are not altogether intact, we could use it to fill in missing
information. GNNs use node classification to predict node attributes, which
in our case could be applied to predict missing passenger information. An
example could be if citizenship information was missing from some of the
passenger data We could use node classification for uncover these citizenship
labels. GNNs also use edge prediction (or link prediction) to uncover hidden
or missing links between nodes. In our case, we could use it to find non-
obvious relationships between the passengers, Examples would be extended
family relationships (cousins on the ship), and business relationships. At the
ship level, if we had similar data for several other large ships, we could
classify the ships themselves and possibly find characteristics that could
portend disaster. Finally, we could encode this graph data, known as
embedding, and use it as additional features in conventional Titanic-focused
machine learning and deep learning solutions. This book will teach how to
practically implement the methods above.

1.2 Goals of the book

GNNs in Action is a book aimed at practitioners who want to deploy graph
neural networks to solve real problems. This could be a machine learning
engineer not familiar with graph data structures, or a graph data scientist who
has not tried machine or deep learning, or software engineers who may not be
familiar with either.

My aim is to enable you to

Assess the suitability of a GNN solution for your problem

Implement a GNN architecture to solve a relevant problem
Understand the tradeoffs in using the different GNN architectures and
graph tools in a production environment
Make clear the limitations of GNNs

As a result, this book is weighted towards implementation using

programming and mainstream data ecosystems. We devote just enough space
on essential theory and concepts, so that the techniques covered can be
sufficiently understood.

I cover the end to end workflow of machine learning - including data loading
and preprocessing, model setup and training, and inference - and apply it to
graphs and GNNs.

I also cover practical guidelines in implementing a GNN solution, including

prototyping and developing applications that will be put into production.

GNNs in Action is a book designed for people to jump quickly into this new
field and start building applications. I hope to offer a book that can reduce the
friction of implementing new technologies by filling in the gaps and
answering key development questions whose answers may have been
heretofore scattered over the internet, or not covered at all. And to do so in a
way that emphasizes approachability rather than high rigor.

In Part 1, we cover fundamentals of graphs, related workflows and data

pipelines, and graph representations. Chapter 2 is devoted to a tour of graphs
and the graph technical ecosystem. Chapter 3 covers the graph data pipeline
from raw data to graph representations. [For readers familiar with graphs, this
material may seem elementary. For those readers, I suggest skimming these
chapters and reviewing topics as needed, then jumping to the chapters on
Graph Neural Networks, which start in Part 2.]

In Part 2, the core of the book, you’ll get introduced to the major types of
GNNs, including Convolutional Networks (GCNs) and GraphSage, Graph
Attention Networks (GATs), Graph Auto-Encoders (GAEs), and GNNs for
Knowledge Graphs

In Part 3, we’ll look at advanced topics, such as customizing GNN

Architectures, Dynamic Graphs, Methods for GNNs at Scale, and Bias in
Graph Data.

Python is the language of choice. At this time, there are several GNN
libraries; our focus will be on Pytorch Geometric, Deep Graph Library, and
Alibaba’s GraphScope. We want this book to be approachable by an audience
with a wide set of hardware constraints, and will try to reflect this in the code
examples. Code will be available in github and colab.

1.3 Why are people excited about GNNs?

In fields that have no lack of hype for new technologies and methods, GNNs
are without hyperbole a leap forward for both the field of deep learning and
for the domain of graph analytics.

In deep learning, previously if one wanted to incorporate features from a

graph into model training, it was done in an indirect way which was not
easily scalable, and could not seamlessly take into account node and edge
properties.

Meanwhile, in the venerable field of graph analysis, several methods have

existed to characterize networks and their elements, many of which are used
in modern graph databases and analytical software [Deo]. However, such
methods have also fallen short in incorporating node and edge properties. In
addition, it has not been possible up to now to develop generalizable models
that could be applied to unseen nodes, edges, or graphs.

Graph neural networks have made strides in remedying these issues, and as a
result are allowing two important types of applications to be developed:

1. Applications based on new and unique data segments and domains; and
2. Applications that enhance traditional machine learning problems in fresh
ways.

Let’s look at examples of both of these applications. Following are two novel
applications of GNNs, and one example of using a GNN in a traditional
application.

Molecular Fingerprints and Interfaces [Sanchez-Lengeling]. In chemistry

and molecular sciences, a prominent problem has been representing
molecules in a general, application-agnostic way, and inferring possible
interfaces between organic molecules, such as proteins. For molecule
representation, we can see that the drawings of molecules that are common in
high school chemistry classes bear resemblance to a graph structure,
consisting of nodes (atoms) and edges (atomic bonds).

Figure 1.4. A molecule. We can see that individual atoms are nodes and the atomic bonds are
edges.

Applying Graph Convolutional Networks to these structures is a nascent field

that promises to outperform traditional ‘fingerprint’ methods, which involve
the creation of features by domain experts to capture a molecule’s properties.

Figure 1.5. A GNN system used to match molecules to odors. The workflow here starts on the left
with a representation of a molecule as a graph. In the middle parts of the figure, this graph
representation is transformed via GNN and MLP layers to a vector representation that can be
trained against odor descriptions on the right. Finally a graph representation of the molecule can
be cast in a space which also corresponds to odors (bottom right).

Mechanical Reasoning [Sanchez-Gonzalez]. From infancy, many living

beings, including humans, have an intuition about mechanics without any
formal training in the subject. Given a bouncing ball, we don’t need pencil
and paper and a set of equations to predict its trajectory. We can innately
forecast where the ball will likely go. We don’t even have to be in the
presence of a physical ball. Given a 2-D snapshot of a bouncing ball, we can
rely on our reasoning to predict in a 3D world where the ball will end up.

Figure 1.6 A graph representation of a mechanical body. The body’s segments are represented as
nodes, and the mechanical forces binding them are edges.
Other documents randomly have
different content
 The forest manager told me he was labouring under the greatest
of disadvantages in his efforts to raise new trees. He said he had to
fight not only the natives, but also the monkeys, baboons, and other
wild animals. The woods are full of monkeys, among them a dog-
faced baboon which grows as big as a ten-year-old boy. This
creature barks like a dog and acts like a devil. It watches the
planting, then sneaks in at night and digs up the trees. If seeds are
put in, it digs them up and bites them in two, and if the trees should
sprout it pulls the sprouts out of the ground and breaks them up and
throws them away. As a result, the nurseries have to be watched all
the time by men with guns in their hands. If the men have no guns
the baboons will jump for the nearest tree and grin from the
branches, only to return to their devastating work as soon as the
watchmen go away. If guns are brought out, the animals realize their
danger and run for their lives. These monkeys also dig up the Indian
corn planted by the Kikuyus, and are said to be far worse than crows
and blackbirds combined.
At one of the stations between Naivasha and the Escarpment I
saw a half-dozen Nandi, including two women. The men were almost
naked, save that they wore cloaks of monkey skins with the fur on
and strips of cowskin about the waist. The women had on waist
cloths and blankets of cowhides tanned with the hair on. These
blankets were fastened over one shoulder, leaving the arms and half
of the breasts bare. The Nandi were walking along the railroad track,
and were closely watched by the station agents, for they are great
thieves, and the British have had trouble with them because they
steal the bolts and rivets which hold the rails to the ties, and even
climb the telegraph poles after the wire. The native men are crazy for
iron. They can use the bolts and rivets for slingshots to brain their
enemies. All the iron they have had in the past has come from
digging up the ore and smelting it, so you can imagine how delightful
it is to a Nandi warrior to pick up a fine, death-dealing iron bolt all
ready for his sling. The Nandi live northwest of Naivasha, on a
plateau which contains iron deposits, and they make a business of
mining and smelting. Since the railroad has been built, they have
come down from time to time and raided the tracks, and the British
have had several little fights with them to drive them off.
 These Nandi are among the bravest of the African natives. They
are much like the Masai, delighting in warfare, and ready to fight at
the least provocation. They are more civilized than the Kikuyus, and
do considerable work in iron and leather. They have cattle, sheep,
and goats, while a few do some farming. Like the Masai, they bleed
their cattle and drink the blood hot, sometimes mixing it with their
porridge. After bleeding, they close the wounds so that the cattle
grow well again. They are good hunters and have large dogs with
which they run down the game, so that it can be killed with spears.
They also trap game by digging wedge-shaped pits and covering
them over with grass. They have donkeys to carry the iron ore from
the mines to their furnaces, where they turn it into pig metal.
I understand that the Nandi live about the same as the other
natives about here. They have circular huts of boards roofed with
thatch. Each hut has a fireplace in the centre on each side of which
is a little bed consisting of a platform of mud built along the wall of
the hut. The people sleep on the mud, using round blocks of wood
for pillows. The children sleep with their parents until they are six
years of age, when they are shoved off into a smaller hut outside
built especially for them. The Nandi believe in witches and medicine
men, and have a sky god to whom they pray every morning and to
whom they sacrifice when times are hard.
Nearly all of these Africans believe in witch doctors. The
Wakamba, whose country I passed through on my way to Nairobi,
not infrequently kill the women of their tribe when they are charged
with witchcraft, and there is a record of something like forty having
been murdered this way within the last few years.
I saw these Wakamba on the Athi plains and in and about Nairobi.
They are tall and fine looking, with woolly hair, rather thick lips, and
almost straight noses.
 CHAPTER XXXV
THE GREAT RIFT VALLEY AND THE MASAI

In the heart of the East African highlands, as far south of the

Mediterranean Sea as New York is distant from Denver, and as far
west of the Indian Ocean as Pittsburgh is west of the Atlantic, I am
writing this chapter. Lake Naivasha, which is spread out before me,
is in the Great Rift Valley, a mighty trough that runs almost north and
south through this part of the continent. This great rift begins, it is
now believed, south of the Zambezi and embraces Lake Rudolf on
the north. Traces of it are to be found even in Palestine. It is
supposed to have been formed by the earth’s folding up after a
stupendous volcanic eruption, which left the craters of Kilimanjaro,
Kenya, and Elgon touching the clouds at altitudes of from fourteen
thousand to nearly twenty thousand feet.
The valley was named by Professor J. W. Gregory, the famous
British geologist, who came out to East Africa in order to explore its
system of valleys and to discover their origin. For many months he
disappeared. There were rumours that he had been killed and cut to
pieces by the Masai. But one day he turned up, looking ill and worn
but triumphant. The results of his trip were published in a book now
historical, “The Great Rift Valley,” from which this huge trough got its
name. To-day one sees everywhere in this part of the country notices
of “Rift Valley” farms or “Rift Valley” hotels. It is still an objective of
scientific explorations and the subject of scientific discussions.
This mighty valley narrows and widens, it rises and falls, and it has
many big lakes. Broadly speaking, all the great lakes of East Africa
are in it or in its spurs. North of here are Lakes Baringo and Rudolf,
and still farther north in Abyssinia is Lake Tsana, the source of the
Blue Nile. As I write I am looking out on Lake Naivasha, a beautiful
sheet of blue water over which white cranes are flying. I can see
zebras and buck feeding not far from the water, and with my glass
can watch the ugly black heads of three hippopotami bobbing up and
down like giant fishing corks upon the surface. The swampy shores
are lined with masses of reeds. Just back of them the ground rises
into rich pastures which are protected from sportsmen by the
reservations allotted to the Uganda Railway and which fairly swarm
with big game.
The weather here is delightful. We are so near the Equator that we
can almost straddle it, but the altitude is such that blankets are
needed at night and it is never excessively hot during the day.
Naivasha is a little higher up in the air than the top of Mt.
Washington. Indeed, the climate of the whole Rift Valley is said to be
suited to white men. This matter is being tested by settlers, for large
tracts of land have been taken up in different places not far from the
railroad, and there are many Englishmen who are going into stock
raising. Near the lake, at Morendat, the government of Kenya Colony
has started an experiment farm and there are big ranches in the
immediate vicinity. There are no tsetse flies here, for even in the
tropics the tsetse is seldom found at an altitude over four thousand
feet. The zebras, which one sees by the hundreds in almost any ride
over the valley, are evidences that horses will thrive. There are also
many ostriches, and in time we may have ostrich farming here as in
South Africa. The average elevation of the lake valley is something
like six thousand feet, and the grass is said to be luxuriant
everywhere.
 Mr. Carpenter, who is five feet eight inches tall, cannot reach more than half way
up the tall stalk of the elephant grass. It has been introduced from Africa into some
of our Southern States and makes a coarse forage crop.
 The Nandi are among the most warlike of the tribes, and before they were
overcome by the British were the terror of more peaceful neighbours. Like the
Masai, they bleed their cattle and drink the blood hot.

This is one of the strongholds of the Masai race, who have always
been noted as warriors and stock raisers. I see them about
Naivasha, and not a few still carry spears and shields. They have
many little towns near by, and their settlements are scattered
throughout the Rift Valley. They live in huts about four feet high, six
feet wide, and nine feet long. The huts, which look like great bake
ovens, are made of branches woven together and plastered with
mud. Sometimes they are smeared over with cow dung, which
material often forms the floors. When it rains, skins are laid over the
roofs to protect them. The houses are usually built in a circle about
an inclosure, in which the cattle are kept at night. The sheep and
goats are allowed to run in and out of the huts. Some of the towns
have fences of thorns around them to keep out the wild beasts.
These Masai are a fierce-looking people. The men are tall and
straight, and walk as though they owned the earth. When they have
their war paint on, they use a decoration of ostrich feathers which
surrounds their faces, and is supposed to carry terror to the souls of
their enemies.
These natives are by no means pure Negroes, but belong to the
Hamitic-negroid or non-Bantu group. Their skins are dark brown,
their noses are often straight, and their lips not very thick. I can’t tell
you whether their hair is woolly or not, for the women shave it close
to the scalp, using razors of iron or glass, and polish their heads with
grease so that they fairly shine in the sun. I understand they pull out
the hair from all parts of their bodies and that even the babies are
shaved. Many of the men carry about tweezers of iron to pull the
hairs from their chins, cheeks, and nostrils, and they keep
themselves shaved until they are old enough to be warriors. This
comes along about the time they reach manhood. They then let the
hair of their heads grow and plait it into pigtails, which they frequently
wear down over the forehead. The head, along with the rest of the
body, is often anointed with oil and red clay. The warrior sometimes
wears a lion’s head and mane in addition to the circle of ostrich
feathers about the face. His arms are a sword and a club. He has a
spear with a very long blade and an oval shield bearing figures which
indicate his clan.
Like the Kikuyus and Nandi, these people buy their wives.
Marriage, however, is not supposed to take place until the Masai
becomes an elder—that is, until he reaches the age of about twenty-
seven or thirty. This is after his fighting days are over and he is ready
to settle down, as it were. The warriors and the young girls of the
tribe live together up to that time in a separate establishment apart
from the rest of the people.
In order to marry, a warrior has to ask permission of the elders of
the tribe. If this is given, he straightway buys his wife. If she is a
good-looking girl she will cost him two cows, two bullocks, two
sheep, and some goatskins. This money goes to the nearest relative
of the woman he has selected, who may lower the price if he will.
Divorces may be had for laziness and bad temper on the part of the
wife; and in such cases a part of the marriage fee is sometimes
returned. A widow cannot marry again. If her husband dies, the relict
goes back to her mother, or to her brother if her mother be dead.
As far as I can learn these Masai girls have a soft snap. They are
required to do nothing until they are married. Before that they play
with the warriors, spending their time in dancing and singing and
loafing about. The unmarried girl often does not do her own cooking.
This condition continues for a long time after marriage and up until
all the babies of the family are fairly well grown. As soon as that is
accomplished, however, the hard-working period begins. Almost all
the hard labour of the tribe is done by the older women, who collect
the firewood, build the mud houses, and gather the cow manure with
which their walls are smeared. When the villages are moved from
place to place, these withered dames take the parts of donkeys and
bullocks in carrying the burdens, and then erect the new huts.
These Masai are a nation of stock raisers and own herds of cattle,
sheep, and goats, which they drive about from pasture to pasture in
the Southern Reservation where the British Government has put
them. The cattle are of the humped variety like the sacred cows of
India, many of them being fat, sleek, and fine looking. Some of the
animals are branded, and not a few have rude bells of iron so they
may be traced if they stray. Most of the cattle are watched by half-
naked boys, who drive them about with sticks. Morning and evening
the cows are brought into the villages to be milked, and nearly every
town of mud huts has its cow houses. The women do the milking.
This is contrary to the custom in some parts of Africa, where it is
thought the cows will go dry if any female touches them. The milk is
caught in gourds which are afterward cleaned with handfuls of burnt
grass. The people always drink their milk fresh, but their method of
cleaning the gourds gives it a smoky flavour. If a calf dies, it is
skinned and stuffed with straw and then placed under the cow’s nose
at milking time, for the natives say the cow will not “let down” her
milk unless the calf is alongside.
The Masai are blood drinkers. Their country has practically no salt,
and I am told that they make up for this lack and keep healthy by
blood drinking.
The people eat but few vegetables and, strangely enough, do not
kill or eat game. They do no farming whatever. Their cooking is
usually done in pots of burnt clay varying from eight to twenty inches
in height. The larger pots are not placed over the fire, but at the side
of it, and are turned around, now and then, in order that they may be
evenly heated.
Much of my information about the Masai comes from Captain
Sidney L. Hinde, who has had a long experience in Africa as an
official, explorer, and lion hunter. He has written some books upon
the Congo and other African countries, and knows much concerning
this part of the world. My talk with Captain Hinde was at Mombasa,
in a beautiful cottage overlooking the Indian Ocean. Upon the floors
were skins of lions and leopards killed by Captain or Mrs. Hinde, and
on the walls were the heads of giraffes, antelope, and gnus shot by
her.
 The Kavirondo wear little in town and less in the country. The tassel hanging
from the waist at the back is the tribal mark of a married woman, while anklets of
telephone wire are the style for both men and women.
 By putting larger and larger objects in the lobes of their ears the natives stretch
them into great loops of flesh, sometimes so long as to be tied under the chin to
keep them from catching in going through the bush.

The evolution of a British colony and how John Bull assumes the
white man’s burden can be read between the lines of my
conversation with these people. Said Captain Hinde:
 “When Mrs. Hinde and I first came into the province the country
was in the same condition it had been in for ages. We found that it
contained about a million people, who lived in little villages, each
containing about ten huts or so. There were no great chiefs. Each
village was independent and almost constantly at war with the
neighbouring villages. The citizens of one settlement knew nothing of
those of the other settlements about. A man dared not venture more
than ten miles from his home, and he had little knowledge of the
country outside that radius. There were no roads whatever excepting
trails which wound this way and that over the land. The only meeting
places were at the markets, which were held at fixed points on
certain days of the week or month. It is a rule throughout Africa that
warfare and fighting must be suspended on market days, and no one
dares bring arms to a market or fight there. If he should engage in
fighting and be killed, his relatives cannot claim blood money.
“When we took possession of the Kenya province we had to fight
our way in. As soon as we had subdued the people, we made them
work at making roads as a penalty for their insurrection. We
connected all the villages by roadways and gave each town so much
to take care of. As a result we now have in that province alone
several hundred miles of good wagon roads each ten feet wide. We
have also made it the law that all roads shall be treated in one
respect like a market place. This means that no native can assault
another while walking upon them and that all feuds must be buried
when travelling over the highways. Many of these roads connect
villages which were formerly at war with each other, and the result of
the law is that they have become peaceful and the citizens can now
pass safely from one town to another. They are really changing their
natures and are going through a process of travel-education. As I
have already said, five years ago they never left home. Now
thousands of them go over our thoroughfares down to the seacoast,
and we have something like eighteen hundred natives of Kenya here
at Mombasa.”
The British have found the Masai such good cattlemen that they
believe they can train them into good grooms for horses. Another
feature of British dealings with the natives is the establishment of
trading posts in the native reserves. Here the Africans are
encouraged to set up little stores of their own. It is hoped that this will
develop wants and help civilize the more backward groups, like the
Masai, until they become as enterprising as the Bagandas and
Kavirondos.
 CHAPTER XXXVI
WHERE THE MEN GO NAKED AND THE WOMEN
WEAR TAILS

Unfurl your fans and take out your kerchiefs to hide your blushes.
We are about to have a stroll among the Kavirondo, who inhabit the
eastern shores of Lake Victoria on the western edge of Kenya
Colony. These people are all more or less naked, and some of the
sights we dare not describe. We have our cameras with us, but our
Postmaster General would not allow some of our films to go through
the mails, and no newspaper would publish certain pictures we take.
We are in the heart of the continent, on the wide Gulf of Kavirondo
on the eastern shore of the second greatest fresh-water lake of the
world. That island-studded sea in front of us is Lake Victoria; and
over there at the northwest, less than a week’s march on foot and
less than two days by the small steamers which ply on the lake, is
Napoleon Gulf, out of which flows that great river, the Nile. With the
glass one may see the hippopotami swimming near the shores of
Kavirondo Bay, while behind us are plains covered with pastures and
spotted with droves of cattle, antelope, and gnu, grazing not far from
the queerly thatched huts of the stark-naked natives.
The plains have a sparse growth of tropical trees, and looking over
them we can catch sight of the hills which steadily rise to the Mau
Escarpment of the Great Rift Valley. Still farther east are the level
highlands of Kenya Colony, the whole extending on and on to
Mombasa and the Indian Ocean, as far distant from Kisumu as
Cleveland is from New York. I have been travelling for days in
coming the five hundred and eighty-four miles which lie between us
and the ocean.
Kisumu, formerly known as Port Florence, is the terminus of the
Uganda Railway, the principal port of Lake Victoria, and quite a
commercial centre. Steamers sail from Kisumu weekly to Uganda
ports and back, and fortnightly round the lake by alternate routes,
i.e., north and south. The trade is greatly increasing, and ivory,
hides, grain, and rubber from Tanganyika Territory, the Upper Congo,
and the lands to the north of the lake are shipped through here to the
coast. The cars come right down to a wooden wharf which extends
well out into the Kavirondo Gulf. On the lake are several small
steamers, brought up here in pieces and put together, which are now
bringing in freight from all parts of this big inland sea.
At the custom house inside an enclosure close to the wharf the
travellers had to pay a fee of fifty cents a package on all parcels
except personal luggage. I was glad we got in before six-thirty, the
closing hour for all custom houses in Uganda ports, for after that if I
were carrying a parcel I should have to slip five rupees to the official
in charge.
Kisumu is just a little tin town in the African wilds, yet there is a
hotel where one can stay quite comfortably until he takes the
steamer for the lake trip. There is an Indian bazaar near the station,
but the post office, the few government buildings, and most of the
residences are built on the hill to get the breeze from the lake. The
Victoria Road and the Connaught Promenade are well laid out. Near
the station there is a cotton ginnery where considerable quantities of
cotton from Uganda are ginned and baled for export. A trail leads
across country from Kisumu to Mumias, forty-eight miles away, and
to Jinja, the source of the Nile.
The European population consists of some soldiers belonging to
the King’s African Rifles, of the government officials, and of some
employees of the railroad. The officials put on great airs. Among the
passengers who came in with me yesterday was a judge who will
settle the disputes among the natives. He was met at the cars by
some soldiers and a gang of convicts in chains. The latter had come
to carry his baggage and other belongings to his galvanized iron
house on the hill and each was dressed in a heavy iron collar with
iron chains extending from it to his wrists and ankles. Nevertheless,
they were able to aid in lifting the boxes and in pushing them off on
trucks, prodded to their work all the while by the guns of the soldiers
on guard.
But let us “take our feet in our hands,” as Uncle Remus says, and
tramp about. Later on we may march off into the country through
which I travelled for about fifty miles on my way here. In the town
itself we may now and then see a man with a blanket wrapped
around him, and the men frequently wear waist cloths behind or in
front. Outside of the town they are stark naked. All have skins of a
dark chocolate brown. They have rather intelligent faces, woolly hair,
and lips and noses like those of a Negro. They belong to the Bantu
family and are among the best formed of the peoples of Africa. Some
one has said that travelling through their country is like walking
through miles of living statuary.
Take these Kavirondo men who have gathered about me just now
as I write. Some of them look as though they might have been cut
from black marble by a sculptor. Look at those three brown bucks at
my left. They are as straight as Michelangelo’s famed statue of
David and about as well formed. See how firmly they stand on their
black feet. Their heads are thrown back and two have burst out
laughing as I turn my camera toward them. With my eye I can follow
the play of all their muscles as they slip beneath those smooth ebony
skins. The Kavirondo seem the perfection of physical manhood. That
nude fellow next me has a coil of wire about his biceps and a pound
of wire on his right wrist. He is smoking a pipe, but it just hangs
between his teeth, which shine out, flashing white as he smiles.
The man next him has two brass rings on each of his black
thumbs, bands of telegraph wire around his wrists, and two wide
coils of wire above and below the biceps of his left arm. He has five
wire bands about his neck, circles of wire under each knee, and
great anklets of twisted wire on each of his feet. As I look I can see
the calloused places where the wire has worn into his instep. There
are worse ones on that third man whose ankles are loaded with
twisted wire. The latter must have several pounds on each leg, and
the wire on the right leg extends from the foot to the middle of the
calf.
 Now look at their heads. The first man has short wool which hugs
the scalp, and the other two have twisted their hair so that it hangs
down about the head like Medusa’s locks.
Stopping for a moment, I ask the men to turn around so I may get
a view from the rear. They are not quite so naked as I had supposed,
for each has an apron of deerskin as big as a lady’s pocket
handkerchief fastened to his waistband behind. The aprons, tanned
with the fur on, are tied to the belts with deerskin straps. As far as
decency goes, they are of no value at all, and they seem to be used
more for ornament than anything else.
Let us train our cameras now on the women. They are by no
means so fine looking as the men, being shorter and not so well
formed. The younger girls are clad in bead waist belts, while the
older ones have each a tassel of fibre tied to a girdle about the waist.
This tassel is fastened just at the small of the back and hangs down
behind. At a short distance it looks like a cow’s tail. I am told that it is
an indispensable article of dress for every married woman, and that
it is improper for a stranger to touch it. Sir Harry Johnston, who once
governed these people, says that even a husband dares not touch
this caudal appendage worn by his wife, and if, by mistake, it is
touched, a goat must be sacrificed or the woman will die from the
insult.
Some of the native women here in Kisumu wear little aprons of
fibre, about six inches long, extending down at the front. I can see
dozens of them so clad all about me, and for a penny or so can get
any of them to pose for my camera. The young girls have no clothes
at all. This is the custom throughout the country. Indeed, farther back
in the interior the fringe aprons are removed, and both sexes are
clad chiefly in wire jewellery of various kinds.
The strangest thing about the nudity of these savages is that they
are absolutely unconscious of any strangeness in it. Such of them as
have not met Europeans do not know they are naked; and a married
woman with her tail of palm fibre feels fully dressed. A traveller tells
how he tried to introduce clothing to a gang of naked young women
whom he met out in the country. He cut up some American sheeting
and gave each girl a piece. They looked at the cloths with interest,
but evidently did not know what to do with them. Thereupon the
white man took a strip and tied it about the waist of one of the party.
Upon this the other girls wrapped their pieces about their waists, but
a moment later took them off, saying: “These are foreign customs
and we do not want them.”
During my stay in the Kavirondo country I have gone out among
the villages and have seen the natives in their homes and at work.
The land is thickly populated. The people are good natured,
enterprising, and quiet. One can go anywhere without danger, and
there is no difficulty in getting photographs of whatever one wants.
I am surprised at the great number of married women. One knows
their status from those sacred tails. The Kavirondo girls marry very
early. They are often betrothed at the age of six years; but in such
case the girl stays with her parents for five or six years afterward.
The parents sell their girls for a price, a good wife being purchasable
for forty hoes, twenty goats, and a cow. In the case of an early
betrothal the suitor pays down part of the fixed sum and the rest in
installments until all is paid. If the father refuses to give up the girl
when the time comes for marriage, the payments having been made,
the suitor organizes a band of his friends, captures her, and carries
her home. A man usually takes his wife from a different village from
that in which he lives. When he comes with his band to the bride’s
village, her gentlemen friends often resist the invasion and fight the
suitor’s party with sticks. At such times the girl screams, but I
understand that she usually allows herself to be captured.
 The witch doctor’s life is safe only so long as the people believe he has power to
break up spells cast upon men or cattle by evil spirits. Most of them come to their
end by violence.
 The British provide for the men who uphold the banner of empire in East Africa
homes that are not only clean but attractive. They have succeeded far beyond any
other nationality as administrators over the millions of primitive blacks.

The Masai, long noted as warriors and cattlemen, live in huts made of branches
woven together and plastered with mud, so that their homes look from a distance
like so many bake-ovens.