Automated Machine Learning for

Business Kai R Larsen

Visit to download the full and correct content document:
https://ebookmass.com/product/automated-machine-learning-for-business-kai-r-larse
n/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Automated Machine Learning for Business Kai R. Larsen

https://ebookmass.com/product/automated-machine-learning-for-
business-kai-r-larsen-2/

Automated Machine Learning for Business R. Larsen

https://ebookmass.com/product/automated-machine-learning-for-
business-r-larsen/

Machine Learning in Microservices: Productionizing

microservices architecture for machine learning
solutions Abouahmed

https://ebookmass.com/product/machine-learning-in-microservices-
productionizing-microservices-architecture-for-machine-learning-
solutions-abouahmed/

Machine Learning for Business Analytics: Concepts,

Techniques and Applications in RapidMiner Galit Shmueli

https://ebookmass.com/product/machine-learning-for-business-
analytics-concepts-techniques-and-applications-in-rapidminer-
galit-shmueli/
Financial Machina: Machine Learning For Finance: The
Quintessential Compendium for Python Machine Learning
For 2024 & Beyond Sampson

https://ebookmass.com/product/financial-machina-machine-learning-
for-finance-the-quintessential-compendium-for-python-machine-
learning-for-2024-beyond-sampson/

Machine Learning for Beginners Aldrich Hill

https://ebookmass.com/product/machine-learning-for-beginners-
aldrich-hill/

Adversarial Robustness for Machine Learning Chen

https://ebookmass.com/product/adversarial-robustness-for-machine-
learning-chen/

Machine Learning for Planetary Science Joern Helbert

https://ebookmass.com/product/machine-learning-for-planetary-
science-joern-helbert/

Machine Learning for Healthcare Applications Sachi

Nandan Mohanty

https://ebookmass.com/product/machine-learning-for-healthcare-
applications-sachi-nandan-mohanty/
Automated Machine Learning for Business
 Automated Machine
Learning for Business
Kai R. Larsen and Daniel S. Becker

1
 3
Oxford University Press is a department of the University of Oxford. It furthers
the University’s objective of excellence in research, scholarship, and education
by publishing worldwide. Oxford is a registered trade mark of Oxford University
Press in the UK and certain other countries.

Published in the United States of America by Oxford University Press

198 Madison Avenue, New York, NY 10016, United States of America.

© Oxford University Press 2021

All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by license, or under terms agreed with the appropriate reproduction
rights organization. Inquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above.

You must not circulate this work in any other form

and you must impose this same condition on any acquirer.

Library of Congress Cataloging-​in-​Publication Data

Names: Larsen, Kai R., author. | Becker, Daniel S., author.
Title: Automated machine learning for business / Kai R. Larsen and Daniel S. Becker.
Description: New York, NY : Oxford University Press, [2021] |
Includes bibliographical references and index.
Identifiers: LCCN 2020049814 (print) | LCCN 2020049815 (ebook) |
ISBN 9780190941659 (hardback) | ISBN 9780190941666 (paperback) |
ISBN 9780190941680 (epub)
Subjects: LCSH: Business planning—Data processing—Textbooks. |
Business planning—Statistical methods—Textbooks. | Machine learning—
Industrial applications—Textbooks. | Decision making—Statistical methods—Textbooks.
Classification: LCC HD30.28 .L3733 2021 (print) |
LCC HD30.28 (ebook) | DDC 658.4/030285631—dc23
LC record available at https://lccn.loc.gov/2020049814
LC ebook record available at https://lccn.loc.gov/2020049815

DOI: 10.1093/​oso/​9780190941659.001.0001

1 3 5 7 9 8 6 4 2
Paperback printed by Sheridan Books, Inc., United States of America
Hardback printed by Bridgeport National Bindery, Inc., United States of America
Preface

According to PricewaterhouseCoopers, there will be an enormous demand for

professionals with the skills you will develop through this book, as the clear majority
of future jobs will be analytics-​enabled (Ampil et al., 2017). Machine learning is at
the core of such jobs and how they are transforming business—​no wonder some
have termed “data scientist” the sexiest job of the twenty-​first century (Davenport &
Patil, 2012). While you may have no desire to become a data scientist, at a minimum,
you must know conceptually what machine learning is, but to thrive you should be
able to use machine learning to make better and faster decisions.
Automated Machine Learning for Business is for these readers (hereafter often re-
ferred to as “analysts”):

• Businesspeople wanting to apply the power of machine learning to learn about

their business environment and extract visualizations allowing the sharing of
their newfound knowledge.
• Businesspeople and developers wanting to learn how to create machine learning
models for automating high-​quality decisions.
• Subject matter experts assigned to a machine learning project. The book will
help you understand the process ahead and allow you to better communicate
with your data science colleagues.
• Students in introductory business analytics or machine learning classes,
whether as part of a business analytics program or a stand-​alone course wholly
or in part focused on machine learning, in either the undergraduate or master-​
level curriculum.
• Machine learning experts with no previous exposure to automated machine
learning or who want to evaluate their machine learning approach against the
industry-​leading processes embedded in DataRobot, the automated machine
learning platform used for this book.

The tools and processes in this book were developed by some of the best data
scientists in the world. Even very successful colleagues with decades of experience
in business analytics and machine learning had useful experiences while testing
material for the book and the DataRobot automated machine learning (AutoML)
platform.
This book is not about artificial intelligence (AI). AI can be thought of as a collec-
tion of machine learning algorithms with a central unit deciding which of the ML
algorithms need to kick in at that time, similar to how different parts of the human
brain specialize in different tasks. Machine learning is in the driving seat as AI is
xii Preface

becoming a reality at blinding speeds. This accelerated pace of AI development is due

to recent improvements in deep learning neural networks as well as other algorithms
that require less fine-​tuning to work. While this book does not cover AI, it may be
one of the gentlest introductions to machine learning, and as such will serve as a
great starting point on your journey toward AI understanding.

Automated Machine Learning (AutoML)

In this book, we teach the machine learning process using a new development in
data science; automated machine learning. AutoML, when implemented properly,
makes machine learning accessible to most people because it removes the need for
years of experience in the most arcane aspects of data science, such as the math,
statistics, and computer science skills required to become a top contender in tradi-
tional machine learning. Anyone trained in the use of AutoML can use it to test their
ideas and support the quality of those ideas during presentations to management
and stakeholder groups. Because the requisite investment is one semester-​long un-
dergraduate course rather than a year in a graduate program, these tools will likely
become a core component of undergraduate programs, and over time, even the high
school curriculum.
It has been our experience that even after taking introductory statistics classes,
only a few are capable of effectively applying a tool like regression to business
problems. On the other hand, most students seem capable of understanding how
to use AutoML, which will generally outperform logistic and linear regression, even
in the cases where regression is the right tool for the problem. If we start with the
assumption that machine learning is in the process of transforming society in the
most radical way since the Industrial Revolution, we can also conclude that devel-
oping undergraduate and graduate business degrees that do not cover this material
will soon be considered akin to educational malpractice. Moreover, several master’s
degree programs in Analytics have started introducing their students to AutoML to
speed their learning and capabilities. When AutoML outperforms whole teams of
experienced data scientists, does it make sense to exclude it from analytics program
training, whether at the undergraduate or graduate level?

A Note to Instructors

The full content of this book may be more than is desired for a class in the core of
an undergraduate curriculum but perhaps not enough content desired for an intro
to data science or business analytics class, as those classes would likely spend more
time on Section III, Acquire and Explore Data, and also cover unsupervised machine
learning. While this book will stick to easy-​to-​use graphical interfaces, the content
of the book is transformable to a class teaching AutoML through a coding approach,
 Preface xiii

primarily with R or Python. Moreover, once your students understand the concepts
covered in this book, they will avoid many pitfalls of machine learning as they move
on to more advanced classes.

Acknowledgments

We are exceedingly grateful to Jeremy Achin and Tom De Godoy, who first built a
system for automating the data science work necessary to compete in Kaggle data
science competitions, a system that over time morphed into the DataRobot platform.
The process taught in this book owes much to their original approach. The book also
could not have become a reality without the support of many world-​leading data
scientists at DataRobot, Inc. We are particularly indebted to Zachary Mayer, Bill
Surrette, João Gomes, Andrew Engel, John Boersma, Raju Penmatcha, Ina Ko, Matt
Sanda, and Ben Solari, who provided feedback on drafts of the book. We are also
grateful to past students Alexander Truesdale, Weston Ballard, and Briana Butler for
support and editing, as well as Alex Truesdale, Briana Butler, Matt Mager, Weston
Ballard, and Spencer Rafii for helping develop the datasets provided with this book.
The following students helped improve the book during its first use in the class-
room at the University of Colorado (in order of helpfulness): Mac Bolak, Gena Welk,
Jackson McCullough, Alan Chen, Stephanie Billett, Robin Silk, Cierra Hubbard,
Meredith Maney, Philip Bubernak, Sarah Sharpe, Meghan McCrory, Megan Thiede,
Pauline Flores, Christine Pracht, Nicole Costello, Tristan Poulsen, Logan Hastings,
Josh Cohn, Alice Haugland, Alex Ward, Meg O’Connell, Sam Silver, Tyler Hennessy,
Daniel Hartman, Anna La, Ben Friedman, Jacob Lamon, Zihan Gao, and August
Ridley.
Finally, but no less important, Kai is grateful to his children, and especially his
youngest daughter, Katrine, who would have received more of his attention had he
not worked on this book.

Book Outline

Section I discusses the ubiquitous use of machine learning in business as well as the
critical need for you to conceptually understand machine learning’s transformative
effect on society. Applications and websites you already use now rely on machine
learning in ways you will never be aware of, as well as the more obvious cases such as
when the car next to you has no driver, or that the driver is watching a video (with or
without holding his hands on the steering wheel). We then discuss automated ma-
chine learning, a new development that makes machine learning accessible to most
businesspeople.
Section II begins with Define Project Objectives, where we specify the business
problem and make sure we have the skill sets needed to succeed. We can then plan
xiv Preface

what to predict. In this section, we will carefully consider the inherent project risks
and rewards, concluding with the truly difficult task of considering whether the pro-
ject is worthwhile or, like most ideas, belongs in the cognitive trash pile.
Section III focuses on how to Acquire and Explore Data. It starts with a consid-
eration of internal versus external data, asking the question of whether the data we
have is the data we need while keeping benefit/​cost considerations in mind. With
all key data collected, the data can be merged into a single table and prepared for
exploratory analysis. In this process, we examine each column in the dataset to de-
termine what data type it is. The most common types of data are either numeric or
categorical. For numeric data, we examine simple statistics, such as mean, median,
standard deviation, minimum value, maximum value. For categorical data, we ex-
amine how many unique categories there are, how many possible values there are in
each category, and within each category, which value is the most common (referred
to as the mode). For both types, we examine their distribution—​are they normally
distributed, left-​or right-​skewed, or perhaps even bimodal? We will also discuss the
potential limitations of categories that have few values.
Traditionally, detailed knowledge of your data has been very important in data sci-
ence, and this will still be true when using AutoML. However, with AutoML, the data
knowledge needed shifts from understanding data to understanding relationships.
A critically important task in data science is to remove target leakage. Target leakage
can occur when we collect a dataset over an extended period such that you have
unrealistic data available at the time of prediction within a model. We will discuss
this in further depth in Chapter 3, but for now, imagine the following: you have ac-
cess to information on a potential customer who just arrived at a website for which
you have contracted to place one of two available ads. You have information on the
previous 100,000 customers who arrived at the website and whether they bought
the product advertised in each ad. You used that information to create a model that
“understands” what leads a customer to buy that product. Your model has an almost
uncanny ability to predict non-​purchasing especially, which means that you will
never place the ad in front of the group predicted not to buy, and you will save your
company quite a bit of money. Unfortunately, after you have convinced your CEO to
place the model into production, you find that not all the features used to create the
model are available when a new visitor arrives, and your model is expected to decide
which ad to place in front of the visitor. It turns out that in your training data, you
had a feature that stated whether visitors clicked on the ad or not. Not surprisingly,
the visitors who did not click the ad also did not buy the product that was only acces-
sible through clicking the ad. You are now forced to scrap the old model and create
a new model before explaining to upper management why your new model is much
less effective (but works). In short, the amount of humble pie eaten by data scientists
who did not remove target leakage before presenting to management could power
many late-​night sessions of feature engineering.
While knowledge of your data is still where you must go to tease out small
improvements in your ability to predict the future, automated machine learning has
 Preface xv

moved the line on the gain from such exercises. The platform we will use to demon-
strate the concepts in this book, DataRobot, does most of the feature engineering
automatically. We have found that for well-​executed automated machine learning,
additional manual feature engineering is much more likely to reduce the perfor-
mance of the algorithms than to improve it. While even the DataRobot platform will
not take away the need for target leakage examination, for AutoML tools embedded
into another platform, such as Salesforce Einstein, target leak data is removed
automatically.
Section IV, Model Data, focuses on creating the machine learning models. Much
of the process of becoming a data scientist emphasizes the aspects of developing
machine learning models, which one does by going through classes on math, sta-
tistics, and programming. Once you have completed your undergraduate degree,
as well as a specialized program in data science or business analytics, you qualify
as a fledgling data scientist, which is admittedly very cool. Then you start building
models and following a process. Often, a machine learning project will take three
to six months before you see your model in production. For example, say that your
university wants you to create a model to predict which students will have trouble
finding a job after achieving their undergraduate degree. While working on the
project, approximately every month you will hear about a new algorithm that is
rumored to be much better than the ones you’ve tried, and you might take a detour
into trying out that algorithm. It has been our experience that often the biggest
problems regarding the accuracy of a machine learning project involve the data
scientist being unaware of a class of algorithms that could have outperformed the
algorithms used.
As part of the modeling process, it is typical to try out different features and per-
haps remove those that are not very predictive or that, due to random variations in
the data, initially seem to offer some predictive ability. Sometimes less is more, and
less is certainly easier to explain to your leadership, to deploy, and to maintain. In
this book, we will focus on the concepts of machine learning rather than the math
and statistics. We will focus on the results of applying an algorithm rather than an
algorithm’s innards. In other words, each algorithm is treated merely regarding its
performance characteristics, each ranked by its performance. Section IV is by far
the most important because here we are working toward an understanding of some-
thing called the confusion matrix (appropriately named). The confusion matrix is a
comparison of whether decisions made by an algorithm were correct. The confusion
matrix gives rise to most of the metrics we use to evaluate models and will also help
drive our business recommendations.
Section V explains how to Interpret and Communicate the model. In this sec-
tion, we move beyond creating and selecting a model to understand what features
drive a target—​in other words, which statistical relationships exist between the
target and the other features. For example, the model you develop will dictate your
management’s trust in the course of action you propose as well as in you. Hence, “we
should invest $100 million in loans to the consumers that the model indicates as
xvi Preface

least likely to default” will require strong faith in either the model or the data scien-
tist who developed the model.
Given that faith in the person develops over time, our focus will be on developing
faith in and understanding of the extent to which the AutoML platform itself not
only finds problems with the model but also clarifies the reason(s) it works when it
does and fails in some cases. Moreover, when specific data turns out to be predictive,
we want to know if this is data to which we can entrust the future of our business.
For example, the purchase of floor protectors (those cheap felt dots placed under
furniture) was in one project found to be as predictive of credit-​worthiness as an
individual’s credit score (Watson, 2014). This feature has instinctive appeal because
one can easily reason that someone who buys floor protectors not only has a wood
floor (more expensive than carpeting) but also cares enough about their property to
protect it from damage. A feature like this could be worth gold if it picked up on vari-
ance unaccounted for by other features associated with credit-​worthiness. However,
a manager would be likely to push back and ask how sure we are that we have data
from all the places a potential customer would buy floor protectors. This same
manager could reasonably worry that the feature only helps us predict the credit-​
worthiness of the top 5% of income earners, a societal segment for which loans come
readily, and competition is fierce. The manager may then worry that over time, word
would get out about the importance of buying $5 floor protectors before applying
for loans, whether one needs them or not. The short story is this: the more of these
unanswered worries your model generates; the slower trust will develop in you as a
data scientist.
Finally, Section VI, Implement, Document and Maintain, sets up a workflow for
using the model to predict new cases. We here describe the stage where we move
the selected model into production. For example, let’s say a consumer’s cell phone
contract is set to expire next week. Before it does, we feed the information on this
consumer into the model, and the model can provide the probability for that con-
sumer to move their service to another cell phone provider. We must then translate
this probability into business rules. For each consumer so scored, we weigh the in-
come from keeping them against the risk of losing them, and then different reten-
tion programs are considered against that risk-​weighted income. It could be that
the consumer will be offered nothing because the risk of losing them is negligible,
or it could be that they will be offered an extra gigabyte of broadband each month,
or perhaps a $200 discount to stay on if the risk-​weighted income is high. Once this
risk-​weighted income is established, the whole process is documented for future re-
producibility and re-​evaluation of rules. The likelihood of a data source changing
becomes a major issue. Just a simple renaming of a column in an Excel spreadsheet
or a database can prevent the whole model from working or lead to significantly
worsened performance, as the model will not recognize the column and all its pre-
dictive capability is lost. We then create a process for model monitoring and main-
tenance. Simply put, over time, the external world modeled by the machine learning
algorithm will change, and if we do not detect this change and retrain the model,
 Preface xvii

the original assumptions that may have made the model profitable will no longer be
true, potentially leading to major losses for the company.
We move forward with the above worries in mind, knowing that we will learn a
framework named the machine learning life cycle to keep us relatively safe by pro-
viding guard-​rails for our work.

Dataset Download

The datasets for the book (described in Appendix A) are available for download at
the following link. The zip file contains one zip file for each dataset (listed as Assets
A.1–​A.8, each referencing the similarly labeled Appendix A datasets). The Zip file
containing the datasets is 131.6MB, and the unzipped files take about 524MB of
space. Download link: https://​www.dropbox.com/​s/​c8qjxdnmclsfsk2/​AutoML_​
DatasetsV1.0.zip?dl=0.

Copyrights

All figures and images not developed for the book, including the cover page image,
are under open usage rights.
Kai Larsen is an Associate Professor of Information Systems at the Leeds School
of Business with a courtesy appointment as an Associate Professor of Information
Science in the College of Media, Communication, and Information at the University
of Colorado, Boulder. In his research, he applies machine learning and natural
language processing (NLP) to address methodological problems in the social and
behavioral sciences. He earned a Ph.D. in Information Science from the Nelson
A. Rockefeller College at SUNY, Albany.
Daniel Becker is a Data Scientist for Google’s Kaggle division. He has broad
data science expertise, with consulting experience for six companies from the
Fortune 100, a second-​place finish in Kaggle's $3 million Heritage Health Prize,
and contributions to the Keras and Tensorflow libraries for deep learning. Dan de-
veloped the training materials for DataRobot University. Dan earned his Ph.D. in
Econometrics from the University of Virginia.
 SECTION I
WHY USE AUTOMATED
MACHINE LEARNING?
1
What Is Machine Learning?

1.1 Why Learn This?

Machine learning is currently at the core of many if not most organizations’

strategies. A recent survey of more than 2,000 organizations’ use of machine
learning and analytics found that these tools are integral for knowing the customer,
streamlining existing operations, and managing risk and compliance. However,
the same organizations were only marginally confident in their analytics-​driven
insights in these areas, including their processes for managing such projects and
evaluating outcomes (KPMG, 2017). In the coming decade, there will be two classes
of organizations: those that use machine learning to transform their capabilities and
those that do not (Davenport, 2006). While barely beyond its inception, the cur-
rent machine learning revolution will affect people and organizations no less than
the Industrial Revolution’s effect on weavers and many other skilled laborers. In the
1700s, weaving required years of experience and extensive manual labor for every
product. This skill set was devalued as the work moved into factories where power
looms vastly improved productivity. Analogously, machine learning will automate
hundreds of millions of jobs that were considered too complex for machines ever to
take over even a decade ago, including driving, flying, painting, programming, and
customer service, as well as many of the jobs previously reserved for humans in the
fields of finance, marketing, operations, accounting, and human resources.
The organizations that use machine learning effectively and survive will most
likely focus on hiring primarily individuals who can help them in their journey of
continuing to derive value from the use of machine learning. The understanding of
how to use algorithms in business will become an essential core competency in the
twenty-​first century. Reading this book and completing any course using it is the
first step in acquiring the skills needed to thrive in this new world. Much has been
made of the need for data scientists, and data scientist salaries certainly support the
premium that industry is placing on such individuals to create and support all the
above applications. A popular meme once decreed that you should “always be your-
self, unless you can be Batman, then always be Batman.” Akin to this: if you can find
a way to be a data scientist, always be a data scientist (especially one as good at his or
her job as Batman is at his), but if you cannot be a data scientist, be the best self you
can be by making sure you understand the machine learning process.
4 Automated Machine Learning for Business

Despite the current doubt within many organizations about their machine
learning capabilities, the odds are that many use one or several technologies with
machine learning built-​in. Examples abound and include fitness bands; digital
assistants like Alexa, Siri, or Cortana; the new machine learning–​powered beds; and
the Nest thermostat, as well as search assistants like Google, where hundreds of ma-
chine learning models contribute to every one of your searches.
You are the subject of machine learning hundreds of times every day. Your so-
cial media posts are analyzed to predict whether you are a psychopath or suffer
from other psychiatric challenges (Garcia & Sikström, 2014), your financial
transactions are examined by the credit-​card companies to detect fraud and money
laundering, and each of us logs into a unique, personalized Amazon Marketplace
tailored to our lives by their highly customizing machine learning algorithms.
Companies have weaved machine learning into everyday life with unseen threads.
For example, machine learning models perform the majority of stock trades, and
even judges contemplating the level of punishment for convicted criminals make
their decisions in consultation with machine learning models. In short, machine
learning algorithms already drive hugely disruptive events for humanity, with major
revolutions still to come.

1.2 Machine Learning Is Everywhere

Everywhere? While there is likely not an ML algorithm at the top of Mount

Everest unless there are also climbers there, there are plenty of machine learning
algorithms working through satellite imagery to put together complete maps not
hampered by clouds (Hu, Sun, Liang & Sun, 2014). These algorithms also predict
poverty through nighttime light intensity (Jean et al., 2016), detect roads (Jean
et al., 2016), detect buildings (Sirmacek & Unsalan, 2011), and generate most of
the 3D structures in Google Earth. If you have tried Google Earth and have floated
through a photo-​realistic Paris or Milan wondering what happened to the cars and
those incredible Italians who still manage to look marvelous despite riding mopeds
to work, an ML algorithm erased them. They simply were too transient and irrele-
vant for Google Earth’s mapping purpose. Zoom in far enough, though, and you’ll
see ghost cars embedded in the asphalt. The important point is that while machine
learning algorithms work wonderfully for most large-​scale problems, if you know
where to look, you’ll find small mistakes. Finding these slips and learning from
them could become one of your defining strengths as an analyst, and you will need
to develop skills in figuring out which features of a dataset are unimportant, both
for the creation of ML models and for determining which results to share with
your boss.
We started to create a complete list of all the areas machine learning has taken
over or will soon take over, but we soon realized it was a meaningless exercise. Just
 What Is Machine Learning? 5

within the last 24 hours of this writing, news has come of Baidu machine learning
that finds missing children in China through facial recognition; Microsoft is de-
veloping machine learning–​powered video search that can locate your keys if
you fill your home with cameras; Facebook announced that their latest language-​
translation algorithm is nine times faster than their competitors’ algorithms;
researchers announced an autonomous robot that uses machine learning to in-
spect bridges for structural damage; and apparently we will soon be able to trans-
late between English and Dolphin (!?). In the meantime, albeit less newsworthy,
dozens if not hundreds of teams of data scientists are engaged in making our lives
better by replacing teen drivers with artificial intelligence that is ten times better
drivers than adolescents right now, with the potential to be 50–​100 fold safer.
Cars have traditionally been 4,000-​pound death machines when left in the care of
people who are sometimes poorly trained, tired, distracted, or under the influence
of medications or drugs. If we leave the driving to machines, car travel will one day
become safer and less stressful.
Currently, machine learning is involved when you ask Alexa/​Cortana/​Siri/​
Google Assistant to search for anything. ML translates your voice to text, and the
Google Assistant uses ML to rank all the pages that contain the keywords you
specified. Increasingly, your very use of the Internet is used to figure you out, some-
times for good purposes, such as offering treatment when you are likely to be de-
pressed (Burns et al., 2011) or recognizing when you are likely to drop out of college
(Baker & Inventado, 2014). Questionable use of ML-​driven identification exists,
such as when Facebook infamously shared insights with a potential customer about
when children were most likely to be emotionally vulnerable and presumably also
more open to certain kinds of ads (Riley, 2017). While such stories are explosive,
the reality is that they remain in the territory of poor PR practices. The sad truth
is that it likely is impossible to conduct large-​scale machine learning to predict
what and when to market a product based on something as emotionally driven
as social media posts and likes without taking advantage of human weaknesses.
Machine learning algorithms will zero in on these weaknesses like a honey badger
on the prowl.
In potentially dubious practices, analysts apply machine learning with a training
set of past consumers that either bought or didn’t buy a given product to predict
new purchasers. Analysts create of a model (a set of weighted relationships) be-
tween the act of buying (vs. not buying) and a set of features (information on the
consumers). These features could be “likes” of a given artist, organization, the
text of their posts, or even behaviors and likes of the person’s friends. If liking an
industrial band (a potential sign of untreated depression) on a Monday morning
predicts later purchase of leather boots, the model will target the people in-
clined toward depressions on the day and time that feeling is the strongest. In
this case, the machine learning will not label these consumers as depressed, or
ever know that they posess a clinical condition, but will still take advantage of
6 Automated Machine Learning for Business

their depression. To evaluate the existence of such unethical shortcuts in models

requires tools that allow careful examination of a model’s “innards.” Being able to
evaluate what happens when a model produces high-​stakes decisions is a critical
reason for learning the content of this book.

1.3 What Is Machine Learning?

Machine learning was defined by Arthur Samuel, a machine learning pioneer, as

“a field of study that gives computers the ability to learn without being explicitly
programmed” (McClendon & Meghanathan, 2015, 3). A slight rewrite on that
suggests that machine learning enables computers to learn to solve a problem by
generalizing from examples (historical data), avoiding the need explicitly to pro-
gram the solution.
There are two types of machine learning: supervised and unsupervised. The
main difference between the two simplifies as supervised machine learning being
a case of the data scientist selecting what they want the machine to learn, whereas
unsupervised machine learning leaves it to the machine to decide what it wants to
learn, with no guarantee that what it learns will be useful to the analyst. This di-
chotomy is a gross simplification, however, as over decades of use, humans have fig-
ured out which unsupervised approaches lead to desirable results.1 An example of
supervised machine learning might be where a model is trained to split people into
two groups: one group, “likely buyers,” and another group, “likely non-​buyers.” The
modeling of relationships in historical data to predict future outcomes is the key
central concept through which machine learning is transforming the world. This is
due to the relative clarity of the idea and the ever-​expanding business potential avail-
able in its implementation. This book will focus exclusively on supervised machine
learning because we as authors believe this is where most of the benefit of machine
learning lies. From here out, we will refer to supervised machine learning as “ma-
chine learning.”
As humans, we create models all the time. Think back to how you learned to
be funny as a child (potentially a life-​saving skill to ensure survival). You were
probably sitting in your high chair being spoon-​fed mashed beans by your father.
Realizing that puréed beans taste horrible, you spit the food back into his face,
resulting in your mom bursting out in laughter. Your brain kicked into predictive
mode, thinking that there must be a relationship between spitting stuff out and
getting a positive reaction (laughter). You try it again and spit the next spoonful
out on the floor. This does not elicit a reaction from mom (no laughter). On the

1 In unsupervised machine learning, models are created without the guidance of historical data on which group

people with certain characteristics belonged to in the past. Examples of unsupervised machine learning is clustering,
factor analysis, and market basket analysis.
 What Is Machine Learning? 7

third spoonful, you decide to replicate the original process, and again your mom
rewards you (laughter). You have now established the relationship between spit-
ting food in the face of a parent and a positive outcome. A few days later, being
alone with your mother, you decide to show once again what a born comedian
you are by spitting food into her face. Much to your surprise, the outcome (no
laughter) indicates that your original model of humor may have been wrong. Over
time, you continued spitting food in adult faces until you had developed a perfect
model of the relationship between this behavior and humor. This is also how ma-
chine learning works, constantly searching for relationships between features and
a target, often as naively as a toddler, but with enough data that the model will out-
perform most human adults.
In machine learning, there is a target (often an outcome) we are trying to under-
stand and predict in future cases. For example, we may want to know the value of a
house given that we know the number of bedrooms, bathrooms, square footage, and
location. Being good at predicting the value of a home (and typically exaggerating it
quite a bit) used to be the main criterion sellers used to select a realtor. Now, websites
such as Zillow.com predict home prices better than any individual realtor ever could.
In business, there are other important targets such as churn (will this customer stop
being a customer?). We also may want to know if a website visitor will click on an
ad for Coach handbags, or whether a job applicant will perform well in your orga-
nization. Healthcare management organizations may want to know which patients
return to the emergency department (ED) within a month of being released, as well
as why they are likely to return. Both the who and the why are important to improve
services and preserve resources. We will return to the example of diabetes patients
released by a hospital ED in this book and use it to understand the aspects of good
patient care. We will also share several datasets covering most business disciplines,
including finance (lending club loans), operations (part backorders), sales (luxury
shoes), reviews (Boston Airbnb), and human resources (HR attrition, two datasets),
as well as some datasets of more general interest, such as student grades and college
starting salaries.
In this book, we will step through a process that will tease out a set of conclusions
about what drives an organizational goal. The process will apply to almost any pro-
cess that has a definable and measurable target. For now, think of machine learning
as analogous to the process you’ve developed through a long life—​the ability to de-
cide whether a meal is going to taste good before the first bite (those mashed beans
served you well in many ways). Every day, we use our already honed sense of taste
to determine which restaurants to go to and which ones to avoid, as well as which of
mom’s meals are worth visiting for, and which ones require a hastily conjured excuse
for escape. Our brains contain an untold number of models that help us predict the
outside world (“I’m really hungry, but this berry looks a lot like the berry that killed
my cousin”). If our forebears had not developed such abilities, we would have long
since died out.
8 Automated Machine Learning for Business

1.4 Data for Machine Learning

More Data Beats a Cleverer Algorithm

ML principle: With the best of feature engineering, results may still not be adequate. Two
options remain: finding/​designing a better algorithm or getting more data. Domingos (2012)
advocates for more data.
AutoML relevance: With AutoML, more data represents the only realistic path to practical
and significant improvements.
  

Let’s use an example to outline how to use machine learning algorithms in busi-
ness. Your CEO just received a report (most likely from you) that 75% of your
eCommerce site customers make only one purchase and that most repeat customers
return within a week. He would like for you to explain what drives a customer to
return. It is now your job to quickly gather as much data as possible starting with
what you know. You are aware that a year ago the organization made large design
changes to the website that changed the traffic patterns quite a bit, so you decide to
start collecting data on customer visits starting a month after that change, allowing
for the stabilization of traffic patterns. Next, you begin processing account creations
from eleven months ago and discard data on accounts created that never led to sales
(a different question that your CEO is sure to ask about later). For each account, you
add a row to a file of customers for your analysis. For each row, you examine whether
the customer made another purchase during the next seven days. If they did, you
code a value of True into a new column called “Retained,” and if not, you code that
customer as False. You stop the addition of customers who created accounts a week
before today’s date to avoid labeling the most recent customers as non-​retained be-
cause they have not had a whole week to return for a second purchase.
Your next task is to gather as much data as possible for both “retained” and “lost”
customers. As such, the specifics of the second purchase are of no interest because
that information is only available about your retained customers. However, the spe-
cific details on the first purchase are important. Likely information that will require
new columns (we call those features) includes:

1. Which site sent them to you? Was it a Google search, Bing search, or did they
come directly to your site?
2. Did this customer come to us through an ad? If so, which ad campaign was it
(presumably each has a unique identifier)?
3. How many different types of products did the customer buy? Was one of the
products a high-​end shoe? Was one of the products a kitchen knife? (Yes,
depending on your company’s product mix, this set of features can become
Another random document with
no related content on Scribd:
would be understood by materialists to indicate some such
phenomenon as the sudden ignition of the star in Cassiopeia, a.d.
1572, and the one in Serpentarius, in 1604, which was noted by
Kepler. But, do the Chaldeans evince in this expression a profounder
philosophy than of our day? Does this change into balls of “pure
divine fire” signify a continuous planetary existence, correspondent
with the spirit-life of man, beyond the awful mystery of death? If
worlds have, as the astronomers tell us, their periods of embryo,
infancy, adolescence, maturity, decadence, and death, may they not,
like man, have their continued existence in a sublimated, ethereal, or
spiritual form? The magians so affirm. They tell us that the fecund
mother Earth is subject to the same laws as every one of her
children. At her appointed time she brings forth all created things; in
the fulness of her days she is gathered to the tomb of worlds. Her
gross, material body slowly parts with its atoms under the inexorable
law which demands their new arrangement in other combinations.
Her own perfected vivifying spirit obeys the eternal attraction which
draws it toward that central spiritual sun from which it was originally
evolved, and which we vaguely know under the name of God.
“And the heaven was visible in seven circles, and the planets
appeared with all their signs, in star-form, and the stars were divided
and numbered with the rulers that were in them, and their revolving
course was bounded with the air, and borne with a circular course,
through the agency of the divine spirit.”[423]
We challenge any one to indicate a single passage in the works of
Hermes which proves him guilty of that crowning absurdity of the
Church of Rome which assumed, upon the geocentric theory of
astronomy, that the heavenly bodies were made for our use and
pleasure, and that it was worth while for the only son of God to
descend upon this cosmic mote and die in expiation for our sins! Mr.
Proctor tells us of a liquid non-permanent shell of uncongealed
matter enclosing a “viscous plastic ocean,” within which “there is
another interior solid globe rotating.” We, on our part, turn to the
Magia Adamica of Eugenius Philalethes, published in 1650, and at
page 12, we find him quoting from Trismegistus in the following
terms: “Hermes affirmeth that in the Beginning the earth was a
quackmire or quivering kind of jelly, it being nothing else but water
congealed by the incubation and heat of the divine spirit; cum adhuc
(sayeth he) Terra tremula esset, Lucente sole compacta esto.”
In the same work, Philalethes, speaking in his quaint, symbolical
way, says, “The earth is invisible ... on my soul it is so, and which is
more, the eye of man never saw the earth, nor can it be seen without
art. To make this element invisible, is the greatest secret in magic ...
as for this fœculent, gross body upon which we walk, it is a compost,
and no earth but it hath earth in it, ... in a word all the elements are
visible but one, namely the earth, and when thou hast attained to so
much perfection as to know why God hath placed the earth in
abscondito,[424] thou hast an excellent figure whereby to know God
Himself, and how He is visible, how invisible.”[425]
Ages before our savants of the nineteenth century came into
existence, a wise man of the Orient thus expressed himself, in
addressing the invisible Deity: “For thy Almighty Hand, that made the
world of formless matter.”[426]
There is much more contained in this language than we are willing
to explain, but we will say that the secret is worth the seeking;
perhaps in this formless matter, the pre-Adamite earth, is contained
a “potency” with which Messrs. Tyndall and Huxley would be glad to
acquaint themselves.
But to descend from universals to particulars, from the ancient
theory of planetary evolution to the evolution of plant and animal life,
as opposed to the theory of special creation, what does Mr. Proctor
call the following language of Hermes but an anticipation of the
modern theory of evolution of species? “When God had filled his
powerful hands with those things which are in nature, and in that
which compasseth nature, then shutting them close again, he said:
‘Receive from me, O holy earth! that art ordained to be the mother of
all, lest thou shouldst want anything;’ when presently opening such
hands as it becomes a God to have, he poured down all that was
necessary to the constitution of things.” Here we have primeval
matter imbued with “the promise and potency of every future form of
life,” and the earth declared to be the predestined mother of
everything that should thenceforth spring from her bosom.
More definite is the language of Marcus Antoninus in his discourse
to himself. “The nature of the universe delights not in anything so
much as to alter all things, and present them under another form.
This is her conceit to play one game and begin another. Matter is
placed before her like a piece of wax and she shapes it to all forms
and figures. Now she makes a bird, then out of the bird a beast—
now a flower, then a frog, and she is pleased with her own magical
performances as men are with their own fancies.”[427]
Before any of our modern teachers thought of evolution, the
ancients taught us, through Hermes, that nothing can be abrupt in
nature; that she never proceeds by jumps and starts, that everything
in her works is slow harmony, and that there is nothing sudden—not
even violent death.
The slow development from preëxisting forms was a doctrine with
the Rosicrucian Illuminati. The Tres Matres showed Hermes the
mysterious progress of their work, before they condescended to
reveal themselves to mediæval alchemists. Now, in the Hermetic
dialect, these three mothers are the symbol of light, heat, and
electricity, or magnetism, the two latter being as convertible as the
whole of the forces or agents which have a place assigned them in
the modern “Force-correlation.” Synesius mentions books of stone
which he found in the temple of Memphis, on which was engraved
the following sentence: “One nature delights in another, one nature
overcomes another, one nature overrules another, and the whole of
them are one.”
The inherent restlessness of matter is embodied in the saying of
Hermes: “Action is the life of Phta;” and Orpheus calls nature
Πολυμήχανος μάτηρ, “the mother that makes many things,” or the
ingenious, the contriving, the inventive mother.
Mr. Proctor says: “All that that is upon and within the earth, all
vegetable forms and all animal forms, our bodies, our brains, are
formed of materials which have been drawn in from those depths of
space surrounding us on all sides.” The Hermetists and the later
Rosicrucians held that all things visible and invisible were produced
by the contention of light with darkness, and that every particle of
matter contains within itself a spark of the divine essence—or light,
spirit—which, through its tendency to free itself from its
entanglement and return to the central source, produced motion in
the particles, and from motion forms were born. Says Hargrave
Jennings, quoting Robertus di Fluctibus: “Thus all minerals in this
spark of life have the rudimentary possibility of plants and growing
organisms; thus all plants have rudimentary sensations which might
(in the ages) enable them to perfect and transmute into locomotive
new creatures, lesser or higher in their grade, or nobler or meaner in
their functions; thus all plants, and all vegetation might pass off (by
side roads) into more distinguished highways as it were, of
independent, completer advance, allowing their original spark of light
to expand and thrill with higher and more vivid force, and to urge
forward with more abounding, informed purpose, all wrought by
planetary influence directed by the unseen spirits (or workers) of the
great original architect.”[428]
Light—the first mentioned in Genesis, is termed by the kabalists,
Sephira, or the Divine Intelligence, the mother of all the Sephiroth,
while the Concealed Wisdom is the father. Light is the first begotten,
and the first emanation of the Supreme, and Light is Life, says the
evangelist. Both are electricity—the life-principle, the anima mundi,
pervading the universe, the electric vivifier of all things. Light is the
great Protean magician, and under the Divine Will of the architect, its
multifarious, omnipotent waves gave birth to every form as well as to
every living being. From its swelling, electric bosom, springs matter
and spirit. Within its beams lie the beginnings of all physical and
chemical action, and of all cosmic and spiritual phenomena; it
vitalizes and disorganizes; it gives life and produces death, and from
its primordial point gradually emerged into existence the myriads of
worlds, visible and invisible celestial bodies. It was at the ray of this
First mother, one in three, that God, according to Plato, “lighted a
fire, which we now call the sun,”[429] and, which is not the cause of
either light or heat, but merely the focus, or, as we might say, the
lens, by which the rays of the primordial light become materialized,
are concentrated upon our solar system, and produce all the
correlations of forces.
So much for the first of Mr. Proctor’s two propositions; now for the
second.
The work which we have been noticing, comprises a series of
twelve essays, of which the last is entitled Thoughts on Astrology.
The author treats the subject with so much more consideration than
is the custom of men of his class, that it is evident he has given it
thoughtful attention. In fact, he goes so far as to say that, “If we
consider the matter aright, we must concede ... that of all the errors
into which men have fallen in their desire to penetrate into futurity,
astrology is the most respectable, we may even say the most
reasonable.”[430]
He admits that “The heavenly bodies do rule the fates of men and
nations in the most unmistakable manner, seeing that without the
controlling and beneficent influences of the chief among those orbs
—the sun—every living creature on the earth must perish.“[431] He
admits, also, the influence of the moon, and sees nothing strange in
the ancients reasoning by analogy, that if two among these heavenly
bodies were thus potent in terrestrial influences, it was ” ... natural
that the other moving bodies known to the ancients, should be
thought to possess also their special powers.”[432] Indeed, the
professor sees nothing unreasonable in their supposition that the
influences exerted by the slower moving planets “might be even
more potent than those of the sun himself.” Mr. Proctor thinks that
the system of astrology “was formed gradually and perhaps
tentatively.” Some influences may have been inferred from observed
events, the fate of this or that king or chief, guiding astrologers in
assigning particular influences to such planetary aspects as were
presented at the time of his nativity. Others may have been invented,
and afterward have found general acceptance, because confirmed
by some curious coincidences.
 A witty joke may sound very prettily, even in a learned treatise,
and the word “coincidence” may be applied to anything we are
unwilling to accept. But a sophism is not a truism; still less is it a
mathematical demonstration, which alone ought to serve as a
beacon—to astronomers, at least. Astrology is a science as infallible
as astronomy itself, with the condition, however, that its interpreters
must be equally infallible; and it is this condition, sine qua non, so
very difficult of realization, that has always proved a stumbling-block
to both. Astrology is to exact astronomy what psychology is to exact
physiology. In astrology and psychology one has to step beyond the
visible world of matter, and enter into the domain of transcendent
spirit. It is the old struggle between the Platonic and Aristotelean
schools, and it is not in our century of Sadducean skepticism that the
former will prevail over the latter. Mr. Proctor, in his professional
capacity, is like the uncharitable person of the Sermon on the Mount,
who is ever ready to attract public attention to the mote in his
despised neighbor’s eye, and overlook the beam in his own. Were
we to record the failures and ridiculous blunders of astronomers, we
are afraid they would outnumber by far those of the astrologers.
Present events fully vindicate Nostradamus, who has been so much
ridiculed by our skeptics. In an old book of prophecies, published in
the fifteenth century (an edition of 1453), we read the following,
among other astrological predictions:[433]

“In twice two hundred years, the Bear

The Crescent will assail;
But if the Cock and Bull unite,
The Bear will not prevail.
In twice ten years again—
Let Islam know and fear—
The Cross shall stand, the Crescent wane,
Dissolve, and disappear.”

In just twice two hundred years from the date of that prophecy, we
had the Crimean war, during which the alliance of the Gallic Cock
and English Bull interfered with the political designs of the Russian
Bear. In 1856 the war was ended, and Turkey, or the Crescent,
closely escaped destruction. In the present year (1876) the most
unexpected events of a political character have just taken place, and
twice ten years have elapsed since peace was proclaimed.
Everything seems to bid fair for a fulfilment of the old prophecy; the
future will tell whether the Moslem Crescent, which seems, indeed,
to be waning, will irrevocably “wane, dissolve, and disappear,” as the
outcome of the present troubles.
In explaining away the heterodox facts which he appears to have
encountered in his pursuit of knowledge, Mr. Proctor is obliged more
than once in his work, to fall back upon these “curious coincidences.”
One of the most curious of these is stated by him in a foot-note
(page 301) as follows: “I do not here dwell on the curious
coincidence—if, indeed, Chaldean astrologers had not discovered
the ring of Saturn—that they showed the god corresponding within a
ring and triple.... Very moderate optical knowledge—such, indeed, as
we may fairly infer from the presence of optical instruments among
Assyrian remains—might have led to the discovery of Saturnal rings
and Jupiter’s moons.... Bel, the Assyrian Jupiter,” he adds, “was
represented sometimes with four star-tipped wings. But it is possible
that these are mere coincidences.”
In short, Mr. Proctor’s theory of coincidence becomes finally more
suggestive of miracle than the facts themselves. For coincidences
our friends the skeptics appear to have an unappeasable appetite.
We have brought sufficient testimony in the preceding chapter to
show that the ancients must have used as good optical instruments
as we have now. Were the instruments in possession of
Nebuchadnezzar of such moderate power, and the knowledge of his
astronomers so very contemptible, when, according to Rawlinson’s
reading of the tiles, the Birs-Nimrud, or temple of Borsippa, had
seven stages, symbolical of the concentric circles of the seven
spheres, each built of tiles and metals to correspond with the color of
the ruling planet of the sphere typified? Is it a coincidence again, that
they should have appropriated to each planet the color which our
latest telescopic discoveries show to be the real one?[434] Or is it
again a coincidence, that Plato should have indicated in the Timæus
his knowledge of the indestructibility of matter, of conservation of
energy, and correlation of forces? “The latest word of modern
philosophy,” says Jowett, “is continuity and development, but to Plato
this is the beginning and foundation of science.”[435]
The radical element of the oldest religions was essentially
sabaistic; and we maintain that their myths and allegories—if once
correctly and thoroughly interpreted, will dovetail with the most exact
astronomical notions of our day. We will say more; there is hardly a
scientific law—whether pertaining to physical astronomy or physical
geography—that could not be easily pointed out in the ingenious
combinations of their fables. They allegorized the most important as
well as the most trifling causes of the celestial motions; the nature of
every phenomenon was personified; and in the mythical biographies
of the Olympic gods and goddesses, one well acquainted with the
latest principles of physics and chemistry can find their causes, inter-
agencies, and mutual relations embodied in the deportment and
course of action of the fickle deities. The atmospheric electricity in its
neutral and latent states is embodied usually in demi-gods and
goddesses, whose scene of action is more limited to earth and who,
in their occasional flights to the higher deific regions, display their
electric tempers always in strict proportion with the increase of
distance from the earth’s surface: the weapons of Hercules and Thor
were never more mortal than when the gods soared into the clouds.
We must bear in mind that before the time when the Olympian
Jupiter was anthropomorphized by the genius of Pheidias into the
Omnipotent God, the Maximus, the God of gods, and thus
abandoned to the adoration of the multitudes, in the earliest and
abstruse science of symbology he embodied in his person and
attributes the whole of the cosmic forces. The Myth was less
metaphysical and complicated, but more truly eloquent as an
expression of natural philosophy. Zeus, the male element of the
creation with Chthonia—Vesta (the earth), and Metis (the water) the
first of the Oceanides (the feminine principles)—was viewed
according to Porphyry and Proclus as the zōŏn-ek-zōōn, the chief of
living beings. In the Orphic theology, the oldest of all, metaphysically
speaking, he represented both the potentia and actus, the
unrevealed cause and the Demiurg, or the active creator as an
emanation from the invisible potency. In the latter demiurgic capacity,
in conjunction with his consorts, we find in him all the mightiest
agents of cosmic evolution—chemical affinity, atmospheric electricity,
attraction, and repulsion.
It is in following his representations in this physical qualification
that we discover how well acquainted were the ancients with all the
doctrines of physical science in their modern development. Later, in
the Pythagorean speculations, Zeus became the metaphysical trinity;
the monad evolving from its invisible self the active cause, effect,
and intelligent will, the whole forming the Tetractis. Still later we find
the earlier Neo-platonists leaving the primal monad aside, on the
ground of its utter incomprehensibleness to human intellect,
speculating merely on the demiurgic triad of this deity as visible and
intelligible in its effects; and thus the metaphysical continuation by
Plotinus, Porphyry, Proclus, and other philosophers of this view of
Zeus the father, Zeus Poseidon, or dunamis, the son and power, and
the spirit or nous. This triad was also accepted as a whole by the
Irenæic school of the second century; the more substantial
difference between the doctrines of the Neo-platonists and the
Christians being merely the forcible amalgamation by the latter of the
incomprehensible monad with its actualized creative trinity.
In his astronomical aspect Zeus-Dionysus has his origin in the
zodiac, the ancient solar year. In Libya he assumed the form of a
ram, and is identical with the Egyptian Amun, who begat Osiris, the
taurian god. Osiris is also a personified emanation of the Father-Sun,
and himself the Sun in Taurus. The Parent-Sun being the Sun in
Aries. As the latter, Jupiter, is in the guise of a ram, and as Jupiter-
Dionysus or Jupiter-Osiris, he is the bull. This animal is, as it is well
known, the symbol of the creative power; moreover the Kabala
explains, through the medium of one of its chief expounders, Simon-
Ben-Iochai,[436] the origin of this strange worship of the bulls and
cows. It is neither Darwin nor Huxley—the founders of the doctrine of
evolution and its necessary complement, the transformation of
species—that can find anything against the rationality of this symbol,
except, perhaps, a natural feeling of uneasiness upon finding that
they were preceded by the ancients even in this particular modern
discovery. Elsewhere, we will give the doctrine of the kabalists as
taught by Simon-Ben-Iochai.
It may be easily proved that from time immemorial Saturn or
Kronos, whose ring, most positively, was discovered by the
Chaldean astrologers, and whose symbolism is no “coincidence,”
was considered the father of Zeus, before the latter became himself
the father of all the gods, and was the highest deity. He was the Bel
or Baal of the Chaldeans, and originally imported among them by the
Akkadians. Rawlinson insists that the latter came from Armenia; but
if so, how can we account for the fact that Bel is but a Babylonian
personification of the Hindu Siva, or Bala, the fire-god, the
omnipotent creative, and at the same time, destroying Deity, in many
senses higher than Brahma himself?
“Zeus,” says an Orphic hymn, “is the first and the last, the head,
and the extremities; from him have proceeded all things. He is a man
and an immortal nymph (male and female element); the soul of all
things; and the principal motor in fire; he is the sun and the moon;
the fountain of the ocean; the demiurgus of the universe; one power,
one God; the mighty creator and governor of the cosmos.
Everything, fire, water, earth, ether, night, the heavens, Metis, the
primeval architecturess (the Sophia of the Gnostics, and the Sephira
of the Kabalists), the beautiful Eros, Cupid, all is included within the
vast dimensions of his glorious body!”[437]
This short hymn of laudation contains within itself the groundwork
of every mythopœic conception. The imagination of the ancients
proved as boundless as the visible manifestations of the Deity itself
which afforded them the themes for their allegories. Still the latter,
exuberant as they seem, never departed from the two principal ideas
which may be ever found running parallel in their sacred imagery; a
strict adherence to the physical as well as moral or spiritual aspect of
natural law. Their metaphysical researches never clashed with
scientific truths, and their religions may be truly termed the psycho-
physiological creeds of the priests and scientists, who built them on
the traditions of the infant-world, such as the unsophisticated minds
of the primitive races received them, and on their own experimental
knowledge, hoary with all the wisdom of the intervening ages.
As the sun, what better image could be found for Jupiter emitting
his golden rays than to personify this emanation in Diana, the all-
illuminating virgin Artemis, whose oldest name was Diktynna, literally
the emitted ray, from the word dikein. The moon is non-luminous,
and it shines only by the reflected light of the sun; hence, the
imagery of his daughter, the goddess of the moon, and herself, Luna,
Astartè, or Diana. As the Cretan Diktynna, she wears a wreath made
of the magic plant diktamnon, or dictamnus, the evergreen shrub
whose contact is said, at the same time, to develop somnambulism
and cure finally of it; and, as Eilithyia and Juno Pronuba, she is the
goddess who presides over births; she is an Æsculapian deity, and
the use of the dictamnus-wreath, in association with the moon,
shows once more the profound observation of the ancients. This
plant is known in botany as possessing strongly sedative properties;
it grows on Mount Dicte, a Cretan mountain, in great abundance; on
the other hand, the moon, according to the best authorities on animal
magnetism, acts upon the juices and ganglionic system, or nerve-
cells, the seat from whence proceed all the nerve-fibres which play
such a prominent part in mesmerization. During childbirth the Cretan
women were covered with this plant, and its roots were administered
as best calculated to soothe acute pain, and allay the irritability so
dangerous at this period. They were placed, moreover, within the
precincts of the temple sacred to the goddess, and, if possible, under
the direct rays of the resplendent daughter of Jupiter—the bright and
warm Eastern moon.
The Hindu Brahmans and Buddhists have complicated theories on
the influence of the sun and moon (the male and female elements),
as containing the negative and positive principles, the opposites of
the magnetic polarity. “The influence of the moon on women is well
known,” write all the old authors on magnetism; and Ennemoser, as
well as Du Potet, confirm the theories of the Hindu seers in every
particular.
 The marked respect paid by the Buddhists to the sapphire-stone—
which was also sacred to Luna, in every other country—may be
found based on something more scientifically exact than a mere
groundless superstition. They ascribed to it a sacred magical power,
which every student of psychological mesmerism will readily
understand, for its polished and deep-blue surface produces
extraordinary somnambulic phenomena. The varied influence of the
prismatic colors on the growth of vegetation, and especially that of
the “blue ray,” has been recognized but recently. The Academicians
quarrelled over the unequal heating power of the prismatic rays until
a series of experimental demonstrations by General Pleasonton,
proved that under the blue ray, the most electric of all, animal and
vegetable growth was increased to a magical proportion. Thus
Amoretti’s investigations of the electric polarity of precious stones
show that the diamond, the garnet, the amethyst, are -E., while the
sapphire is +E.[438] Thus, we are enabled to show that the latest
experiments of science only corroborate that which was known to the
Hindu sages before any of the modern academies were founded. An
old Hindu legend says that Brahma-Prajapâti, having fallen in love
with his own daughter, Ushâs (Heaven, sometimes the Dawn also),
assumed the form of a buck (ris’ya) and Ushâs that of a female deer
(rôhit) and thus committed the first sin.[439] Upon seeing such a
desecration, the gods felt so terrified, that uniting their most fearful-
looking bodies—each god possessing as many bodies as he desires
—they produced Bhûtavan (the spirit of evil), who was created by
them on purpose to destroy the incarnation of the first sin committed
by the Brahma himself. Upon seeing this, Brahma-
Hiranyagarbha[440] repented bitterly and began repeating the
Mantras, or prayers of purification, and, in his grief, dropped on earth
a tear, the hottest that ever fell from an eye; and from it was formed
the first sapphire.
This half-sacred, half-popular legend shows that the Hindus knew
which was the most electric of all the prismatic colors; moreover, the
particular influence of the sapphire-stone was as well defined as that
of all the other minerals. Orpheus teaches how it is possible to affect
a whole audience by means of a lodestone; Pythagoras pays a
particular attention to the color and nature of precious stones; while
Apollonius of Tyana imparts to his disciples the secret virtues of
each, and changes his jewelled rings daily, using a particular stone
for every day of the month and according to the laws of judicial
astrology. The Buddhists assert that the sapphire produces peace of
mind, equanimity, and chases all evil thoughts by establishing a
healthy circulation in man. So does an electric battery, with its well-
directed fluid, say our electricians. “The sapphire,” say the
Buddhists, “will open barred doors and dwellings (for the spirit of
man); it produces a desire for prayer, and brings with it more peace
than any other gem; but he who would wear it must lead a pure and
holy life.”[441]
Diana-Luna is the daughter of Zeus by Proserpina, who represents
the Earth in her active labor, and, according to Hesiod, as Diana
Eilythia-Lucina she is Juno’s daughter. But Juno, devoured by
Kronos or Saturn, and restored back to life by the Oceanid Metis, is
also known as the Earth. Saturn, as the evolution of Time, swallows
the earth in one of the ante-historical cataclysms, and it is only when
Metis (the waters) by retreating in her many beds, frees the
continent, that Juno is said to be restored to her first shape. The idea
is expressed in the 9th and 10th verses of the first chapter of
Genesis. In the frequent matrimonial quarrels between Juno and
Jupiter, Diana is always represented as turning her back on her
mother and smiling upon her father, though she chides him for his
numerous frolics. The Thessalian magicians are said to have been
obliged, during such eclipses, to draw her attention to the earth by
the power of their spells and incantations, and the Babylonian
astrologers and magi never desisted in their spells until they brought
about a reconciliation between the irritated couple, after which Juno
“radiantly smiled on the bright goddess” Diana, who, encircling her
brow with her crescent, returned to her hunting-place in the
mountains.
It seems to us that the fable illustrates the different phases of the
moon. We, the inhabitants of the earth, never see but one-half of our
bright satellite, who thus turns her back to her mother Juno. The sun,
the moon, and the earth are constantly changing positions with
relation to each other. With the new moon there is constantly a
change of weather; and sometimes the wind and storms may well
suggest a quarrel between the sun and earth, especially when the
former is concealed by grumbling thunder-clouds. Furthermore, the
new moon, when her dark side is turned toward us, is invisible; and it
is only after a reconciliation between the sun and the earth, that a
bright crescent becomes visible on the side nearest to the sun,
though this time Luna is not illuminated by sunlight directly received,
but by sunlight reflected from the earth to the moon, and by her
reflected back to us. Hence, the Chaldean astrologers and the
magicians of Thessaly, who probably watched and determined as
accurately as a Babinet the course of the celestial bodies, were said
by their enchantments to force the moon to descend on earth, i.e., to
show her crescent, which she could do but after receiving the
“radiant smile” from her mother-earth, who put it on after the conjugal
reconciliation. Diana-Luna, having adorned her head with her
crescent, returns back to hunt in her mountains.
As to calling in question the intrinsic knowledge of the ancients on
the ground of their “superstitious deductions from natural
phenomena,” it is as appropriate as it would be if, five hundred years
hence, our descendents should regard the pupils of Professor
Balfour Stewart as ancient ignoramuses, and himself a shallow
philosopher. If modern science, in the person of this gentleman, can
condescend to make experiments to determine whether the
appearance of the spots on the sun’s surface is in any way
connected with the potatoe disease, and finds it is; and that,
moreover, “the earth is very seriously affected by what takes place in
the sun,”[442] why should the ancient astrologers be held up as either
fools or arrant knaves? There is the same relation between natural
and judicial or judiciary astrology, as between physiology and
psychology, the physical and the moral. If in later centuries these
sciences were degraded into charlatanry by some money-making
impostors, is it just to extend the accusation to those mighty men of
old who, by their persevering studies and holy lives, bestowed an
immortal name upon Chaldea and Babylonia? Surely those who are
now found to have made correct astronomical observations ranging
back to “within 100 years from the flood,” from the top observatory of
the “cloud-encompassed Bel,” as Prof. Draper has it, can hardly be
considered impostors. If their mode of impressing upon the popular
minds the great astronomical truths differed from the “system of
education” of our present century and appears ridiculous to some,
the question still remains unanswered: which of the two systems was
the best? With them science went hand in hand with religion, and the
idea of God was inseparable from that of his works. And while in the
present century there is not one person out of ten thousand who
knows, if he ever knew the fact at all, that the planet Uranus is next
to Saturn, and revolves about the sun in eighty-four years; and that
Saturn is next to Jupiter, and takes twenty-nine and a half years to
make one complete revolution in its orbit; while Jupiter performs his
revolution in twelve years; the uneducated masses of Babylon and
Greece, having impressed on their minds that Uranus was the father
of Saturn, and Saturn that of Jupiter, considering them furthermore
deities as well as all their satellites and attendants, we may perhaps
infer from it, that while Europeans only discovered Uranus in 1781, a
curious coincidence is to be noticed in the above myths.
We have but to open the most common book on astrology, and
compare the descriptions embraced in the Fable of the Twelve
Houses with the most modern discoveries of science as to the nature
of the planets and the elements in each star, to see that without any
spectroscope the ancients were perfectly well acquainted with the
same. Unless the fact is again regarded as “a coincidence,” we can
learn, to a certain extent, of the degree of the solar heat, light, and
nature of the planets by simply studying their symbolic
representations in the Olympic gods, and the twelve signs of the
zodiac, to each of which in astrology is attributed a particular quality.
If the goddesses of our own planet vary in no particular from other
gods and goddesses, but all have a like physical nature, does not
this imply that the sentinels who watched from the top of Bel’s tower,
by day as well as by night, holding communion with the euhemerized
deities, had remarked, before ourselves, the physical unity of the
universe and the fact that the planets above are made of precisely
the same chemical elements as our own. The sun in Aries, Jupiter, is
shown in astrology as a masculine, diurnal, cardinal, equinoctial,
easterly sign, hot and dry, and answers perfectly to the character
attributed to the fickle “Father of the gods.” When angry Zeus-Akrios
snatches from his fiery belt the thunderbolts which he hurls forth
from heaven, he rends the clouds and descends as Jupiter Pluvius in
torrents of rain. He is the greatest and highest of gods, and his
movements are as rapid as lightning itself. The planet Jupiter is
known to revolve on its axis so rapidly that the point of its equator
turns at the rate of 450 miles a minute. An immense excess of
centrifugal force at the equator is believed to have caused the planet
to become extremely flattened at the poles; and in Crete the
personified god Jupiter was represented without ears. The planet
Jupiter’s disk is crossed by dark belts; varying in breadth, they
appear to be connected with its rotation on its axis, and are
produced by disturbances in its atmosphere. The face of Father
Zeus, says Hesiod, became spotted with rage when he beheld the
Titans ready to rebel.
In Mr. Proctor’s book, astronomers seem especially doomed by
Providence to encounter all kinds of curious “coincidences,” for he
gives us many cases out of the “multitude,” and even of the
“thousands of facts [sic].” To this list we may add the army of
Egyptologists and archæologists who of late have been the chosen
pets of the capricious Dame Chance, who, moreover, generally
selects “well-to-do Arabs” and other Eastern gentlemen, to play the
part of benevolent genii to Oriental scholars in difficulties. Professor
Ebers is one of the latest favored ones. It is a well-known fact, that
whenever Champollion needed important links, he fell in with them in
the most various and unexpected ways.
Voltaire, the greatest of “infidels” of the eighteenth century, used to
say, that if there were no God, people would have to invent one.
Volney, another “materialist,” nowhere throughout his numerous
writings denies the existence of God. On the contrary, he plainly
asserts several times that the universe is the work of the “All-wise,”
and is convinced that there is a Supreme Agent, a universal and
identical Artificer, designated by the name of God.[443] Voltaire
becomes, toward the end of his life, Pythagorical, and concludes by
saying: “I have consumed forty years of my pilgrimage ... seeking the
philosopher’s stone called truth. I have consulted all the adepts of
antiquity, Epicurus and Augustine, Plato and Malebranche, and I still
remain in ignorance.... All that I have been able to obtain by
comparing and combining the system of Plato, of the tutor of
Alexander, Pythagoras, and the Oriental, is this: Chance is a word
void of sense. The world is arranged according to mathematical
laws.”[444]
It is pertinent for us to suggest that Mr. Proctor’s stumbling-block is
that which trips the feet of all materialistic scientists, whose views he
but repeats; he confounds the physical and spiritual operations of
nature. His very theory of the probable inductive reasoning of the
ancients as to the subtile influences of the more remote planets, by
comparison with the familiar and potent effects of the sun and moon
upon our earth, shows the drift of his mind. Because science affirms
that the sun imparts physical heat and light to us, and the moon
affects the tides, he thinks that the ancients must have regarded the
other heavenly bodies as exerting the same kind of influence upon
us physically, and indirectly upon our fortunes.[445] And here we
must permit ourselves a digression.
How the ancients regarded the heavenly bodies is very hard to
determine, for one unacquainted with the esoteric explanation of
their doctrines. While philology and comparative theology have
begun the arduous work of analysis, they have as yet arrived at
meagre results. The allegorical form of speech has often led our
commentators so far astray, that they have confounded causes with
effects, and vice versa. In the baffling phenomenon of force-
correlation, even our greatest scientists would find it very hard to
explain which of these forces is the cause, and which the effect,
since each may be both by turns, and convertible. Thus, if we should
inquire of the physicists, “Is it light which generates heat, or the latter
which produces light?” we would in all probability be answered that it
is certainly light which creates heat. Very well; but how? did the great
Artificer first produce light, or did He first construct the sun, which is
said to be the sole dispenser of light, and, consequently, heat?
These questions may appear at first glance indicative of ignorance;
but, perhaps, if we ponder them deeply, they will assume another
appearance. In Genesis, the “Lord” first creates light, and three days
and three nights are alleged to pass away before He creates the
sun, the moon, and the stars. This gross blunder against exact
science has created much merriment among materialists. And they
certainly would be warranted in laughing, if their doctrine that our
light and heat are derived from the sun were unassailable. Until
recently, nothing has happened to upset this theory, which, for lack
of a better one, according to the expression of a preacher, “reigns
sovereign in the Empire of Hypothesis.” The ancient sun-
worshippers regarded the Great Spirit as a nature-god, identical with
nature, and the sun as the deity, “in whom the Lord of life dwells.”
Gama is the sun, according to the Hindu theology, and “The sun is
the source of the souls and of all life.”[446] Agni, the “Divine Fire,” the
deity of the Hindu, is the sun,[447] for the fire and sun are the same.
Ormazd is light, the Sun-God, or the Life-giver. In the Hindu
philosophy, “The souls issue from the soul of the world, and return to
it as sparks to the fire.”[448] But, in another place, it is said that “The
Sun is the soul of all things; all has proceeded out of it, and will
return to it,”[449] which shows that the sun is meant allegorically here,
and refers to the central, invisible sun, GOD, whose first
manifestation was Sephira, the emanation of En-Soph—Light, in
short.
“And I looked, and behold, a whirlwind came out of the north, a
great cloud, and a fire infolding itself, and a brightness was about it,”
says Ezekiel (i. 4, 22, etc.), “ ... and the likeness of a throne ... and
as the appearance of a man above upon it ... and I saw as it were
the appearance of fire and it had brightness round about it.” And
Daniel speaks of the “ancient of days,” the kabalistic En-Soph,
whose throne was “the fiery flame, his wheels burning fire.... A fiery
stream issued and came forth from before him.”[450] Like the Pagan
Saturn, who had his castle of flame in the seventh heaven, the
Jewish Jehovah had his “castle of fire over the seventh
heavens.”[451]
 If the limited space of the present work would permit we might
easily show that none of the ancients, the sun-worshippers included,
regarded our visible sun otherwise than as an emblem of their
metaphysical invisible central sun-god. Moreover, they did not
believe what our modern science teaches us, namely, that light and
heat proceed from our sun, and that it is this planet which imparts all
life to our visible nature. “His radiance is undecaying,” says the Rig-
Veda, “the intensely-shining, all-pervading, unceasing, undecaying
rays of Agni desist not, neither night nor day.” This evidently related
to the spiritual, central sun, whose rays are all-pervading and
unceasing, the eternal and boundless life-giver. He the Point; the
centre (which is everywhere) of the circle (which is nowhere), the
ethereal, spiritual fire, the soul and spirit of the all-pervading,
mysterious ether; the despair and puzzle of the materialist, who will
some day find that that which causes the numberless cosmic forces
to manifest themselves in eternal correlation is but a divine
electricity, or rather galvanism, and that the sun is but one of the
myriad magnets disseminated through space—a reflector—as
General Pleasonton has it. That the sun has no more heat in it than
the moon or the space-crowding host of sparkling stars. That there is
no gravitation in the Newtonian sense,[452] but only magnetic
attraction and repulsion; and that it is by their magnetism that the
planets of the solar system have their motions regulated in their
respective orbits by the still more powerful magnetism of the sun, not
by their weight or gravitation. This and much more they may learn;
but, until then we must be content with being merely laughed at,
instead of being burned alive for impiety, or shut up in an insane
asylum.
The laws of Manu are the doctrines of Plato, Philo, Zoroaster,
Pythagoras, and of the Kabala. The esoterism of every religion may
be solved by the latter. The kabalistic doctrine of the allegorical
Father and Son, or Πατηρ and Λογος is identical with the
groundwork of Buddhism. Moses could not reveal to the multitude
the sublime secrets of religious speculation, nor the cosmogony of
the universe; the whole resting upon the Hindu Illusion, a clever
mask veiling the Sanctum Sanctorum, and which has misled so
many theological commentators.[453]
The kabalistic heresies receive an unexpected support in the
heterodox theories of General Pleasonton. According to his opinions
(which he supports on far more unimpeachable facts than orthodox
scientists theirs) the space between the sun and the earth must be
filled with a material medium, which, so far as we can judge from his
description, answers to our kabalistic astral light. The passage of
light through this must produce enormous friction. Friction generates
electricity, and it is this electricity and its correlative magnetism which
forms those tremendous forces of nature that produce in, on, and
about our planet the various changes which we everywhere
encounter. He proves that terrestrial heat cannot be directly derived
from the sun, for heat ascends. The force by which heat is effected is
a repellent one, he says, and as it is associated with positive
electricity, it is attracted to the upper atmosphere by its negative
electricity, always associated with cold, which is opposed to positive
electricity. He strengthens his position by showing that the earth,
which when covered with snow cannot be affected by the sun’s rays,
is warmest where the snow is deepest. This he explains upon the
theory that the radiation of heat from the interior of the earth,
positively electrified, meeting at the surface of the earth with the
snow in contact with it, negatively electrified, produces the heat.
Thus he shows that it is not at all to the sun that we are indebted
for light and heat; that light is a creation sui generis, which sprung
into existence at the instant when the Deity willed, and uttered the
fiat: “Let there be light;” and that it is this independent material agent
which produces heat by friction, on account of its enormous and
incessant velocity. In short, it is the first kabalistic emanation to
which General Pleasonton introduces us, that Sephira or divine
Intelligence (the female principle), which, in unity with En-Soph, or
divine wisdom (male principle) produced every thing visible and
invisible. He laughs at the current theory of the incandescence of the
sun and its gaseous substance. The reflection from the photosphere
of the sun, he says, passing through planetary and stellar spaces,
must have thus created a vast amount of electricity and magnetism.