Data science is the discipline of extracting actionable knowledge from data of various forms. The name data science emerged quite recently--it was invented by DJ Patil and Jeff Hammerbacher and popularized in the article Data Scientist: The Sexiest Job of the 21st Century in 2012. But the discipline itself had existed before for quite a while and previously was known by other names such as data mining or predictive analytics. Data science, like its predecessors, is built on statistics and machine learning algorithms for knowledge extraction and model building.

The science part of the term data science is no coincidence--if we look up science, its definition can be summarized to systematic organization of knowledge in terms testable explanations and predictions. This is exactly what data scientists do, by extracting patterns from available data, they can make predictions about future unseen data, and they make sure the predictions are validated beforehand. 

Nowadays, data science is used across many fields, including (but not limited to):

• Banking: Risk management (for example, credit scoring), fraud detection, trading
• Insurance: Claims management (for example, accelerating claim approval), risk and losses estimation, also fraud detection
• Health care: Predicting diseases (such as strokes, diabetes, cancer) and relapses
• Retail and e-commerce: Market basket analysis (identifying product that go well together), recommendation engines, product categorization, and personalized searches

This book covers the following practical use cases:

• Predicting whether an URL is likely to appear on the first page of a search engine
• Predicting how fast an operation will be completed given the hardware specifications
• Ranking text documents for a search engine
• Checking whether there is a cat or a dog on a picture
• Recommending friends in a social network
• Processing large-scale textual data on a cluster of computers

In all these cases, we will use data science to learn from data and use the learned knowledge to solve a particular business problem.

We will also use a running example throughout the book, building a search engine. We will use it to illustrate many data science concepts such as, supervised machine learning, dimensionality reduction, text mining, and learning to rank models. 

Machine learning is a part of computer science, and it is at the core of data science. The data itself, especially in big volumes, is hardly useful, but inside it hides highly valuable patterns. With the help of machine learning, we can recognize these hidden patterns, extract them, and then apply the learned information to the new unseen items. 

For example, given the image of an animal, a machine learning algorithm can say whether the picture is a dog or a cat; or, given the history of a bank client, it will say how likely the client is to default, that is, to fail to pay the debt.

Often, machine learning models are seen as black boxes that take in a data point and output a prediction for it. In this book, we will look at what is inside these black boxes and see how and when it is best to use them.

The typical problems that machine learning solves can be categorized in the following groups:

• Supervised learning: For each data point, we have a label--extra information that describes the outcome that we want to learn. In the cats versus dogs case, the data point is an image of the animal; the label describes whether it's a dog or a cat.
• Unsupervised learning: We only have raw data points and no label information is available. For example, we have a collection of e-mails and we would like to group them based on how similar they are. There is no explicit label associated with the e-mails, which makes this problem unsupervised.
• Semi-supervised learning: Labels are given only for a part of the data.
• Reinforcement learning: Instead of labels, we have a reward; something the model gets by interacting with the environment it runs in. Based on the reward, it can adapt and maximize it. For example, a model that learns how to play chess gets a positive reward each time it eats a figure of the opponent, and gets a negative reward each time it loses a figure; and the reward is proportional to the value of the figure. 

As we discussed previously, for supervised learning we have some information attached to each data point, the label, and we can train a model to use it and to learn from it. For example, if we want to build a model that tells us whether there is a dog or a cat on a picture, then the picture is the data point and the information whether it is a dog or a cat is the label. Another example is predicting the price of a house--the description of a house is the data point, and the price is the label. 

We can group the algorithms of supervised learning into classification and regression algorithms based on the nature of this information.

In classification problems, the labels come from some fixed finite set of classes, such as {cat, dog}, {default, not default}, or {office, food, entertainment, home}. Depending on the number of classes, the classification problem can be binary (only two possible classes) or multi-class (several classes).

Examples of classification algorithms are Naive Bayes, logistic regression, perceptron, Support Vector Machine (SVM), and many others. We will discuss classification algorithms in more detail in the first part of Chapter 4, Supervised Learning - Classification and Regression.

In regression problems, the labels are real numbers. For example, a person can have a salary in the range from $0 per year to several billions per year. Hence, predicting the salary is a regression problem.

Examples of regression algorithms are linear regression, LASSO, Support Vector Regression (SVR), and others. These algorithms will be described in more detail in the second part of Chapter 4, Supervised Learning - Classification and Regression.

Some of the supervised learning methods are universal and can be applied to both classification and regression problems. For example, decision trees, random forest, and other tree-based methods can tackle both types. We will discuss one such algorithm, gradient boosting machines in Chapter 7, Extreme Gradient Boosting.

Neural networks can also deal with both classification and regression problems, and we will talk about them in Chapter 8, Deep Learning with DeepLearning4J.

Unsupervised learning covers the cases where we have no labels available, but still want to find some patterns hidden in the data. There are several types of unsupervised learning, and we will look into cluster analysis, or clustering and unsupervised dimensionality reduction. 

Typically, when people talk about unsupervised learning, they talk about cluster analysis or clustering. A cluster analysis algorithm takes a set of data points and tries to categorize them into groups such that similar items belong to the same group, and different items do not. There are many ways where it can be used, for example, in customer segmentation or text categorization.

Customer segmentation is an example of clustering. Given some description of customers, we try to put them into groups such that the customers in one group have similar profiles and behave in a similar way. This information can be used to understand what do the people in these groups want, and this can be used to target them with better advertisements and other promotional messages.

Another example is text categorization. Given a collection of texts, we would like to find common topics among these texts and arrange the texts according to these topics. For example, given a set of complaints in an e-commerce store, we may want to put ones that talk about similar things together, and this should help the users of the system navigate through the complaints easier.

Examples of cluster analysis algorithms are hierarchical clustering, k-means, density-based spatial clustering of applications with noise (DBSCAN), and many others. We will talk about clustering in detail in the first part of Chapter 5, Unsupervised Learning - Clustering and Dimensionality Reduction.

Another group of unsupervised learning algorithms is dimensionality reduction algorithms. This group of algorithms compresses the dataset, keeping only the most useful information. If our dataset has too much information, it can be hard for a machine learning algorithm to use all of it at the same time. It may just take too long for the algorithm to process all the data and we would like to compress the data, so processing it takes less time. 

There are multiple algorithms that can reduce the dimensionality of the data, including Principal Component Analysis (PCA), Locally linear embedding, and t-SNE. All these algorithms are examples of unsupervised dimensionality reduction techniques.

Not all dimensionality reduction algorithms are unsupervised; some of them can use labels to reduce the dimensionality better. For example, many feature selection algorithms rely on labels to see what features are useful and what are not. 

We will talk more about this in Chapter 5, Unsupervised Learning - Clustering and Dimensionality Reduction.

Processing natural language texts is very complex, they are not very well structured and require a lot of cleaning and normalizing. Yet the amount of textual information around us is tremendous: a lot of text data is generated every minute, and it is very hard to retrieve useful information from them. Using data science and machine learning is very helpful for text problems as well; they allow us to find the right text, process it, and extract the valuable bits of information.

There are multiple ways we can use the text information. One example is information retrieval, or, simply, text search--given a user query and a collection of documents, we want to find what are the most relevant documents in the corpus with respect to the query, and present them to the user. Other applications include sentiment analysis--predicting whether a product review is positive, neutral or negative, or grouping the reviews according to how they talk about the products. 

We will talk more about information retrieval, Natural Language Processing (NLP) and working with texts in Chapter 6, Working with Text - Natural Language Processing and Information Retrieval. Additionally, we will see how to process large amounts of text data in Chapter 9, Scaling Data Science.  

The methods we can use for machine learning and data science are very important. What is equally important is the the way we create them and then put them to use in production systems. Data science process models help us make it more organized and systematic, which is why we will talk about them next.

Applying data science is much more than just selecting a suitable machine learning algorithm and using it on the data. It is always good to keep in mind that machine learning is only a small part of the project; there are other parts such as understanding the problem, collecting the data, testing the solution and deploying to the production.

When working on any project, not just data science ones, it is beneficial to break it down into smaller manageable pieces and complete them one-by-one. For data science, there are best practices that describe how to do it the best way, and they are called process models. There are multiple models, including CRISP-DM and OSEMN.

In this chapter, CRISP-DM is explained as Obtain, Scrub, Explore, Model, and iNterpret (OSEMN), which is more suitable for data analysis tasks and addresses many important steps to a lesser extent.

Cross Industry Standard Process for Data Mining (CRISP-DM) is a process methodology for developing data mining applications. It was created before the term data science became popular, it's reliable and time-tested by several generations of analytics. These practices are still useful nowadays and describe the high-level steps of any analytical project quite well. 

The CRISP-DM methodology breaks down a project into the following steps:

• Business understanding
• Data understanding
• Data preparation
• Modeling
• Evaluation
• Deployment

The methodology itself defines much more than just these steps, but typically knowing what the steps are and what happens at each step is enough for a successful data science project. Let's look at each of these steps separately.

The first step is Business Understanding. This step aims at learning what kinds of problems the business has and what they want to achieve by solving these problems. To be successful, a data science application must be useful for the business. The result of this step is the formulation of a problem which we want to solve and what is the desired outcome of the project.

The second step is Data Understanding. In this step, we try to find out what data can be used to solve the problem. We also need to find out if we already have the data; if not, we need to think how we can we get it. Depending on what data we find (or do not find), we may want to alter the original goal.

When the data is collected, we need to explore it. The process of reviewing the data is often called Exploratory Data Analysis and it is an integral part of any data science project. it helps to understand the processes that created the data, and can already suggest approaches for tackling the problem. The result of this step is the knowledge about which data sources are needed to solve the problem. We will talk more about this step in Chapter 3, Exploratory Data Analysis.

The third step of CRISP-DM is Data Preparation. For a dataset to be useful, it needs to be cleaned and transformed to a tabular form. The tabular form means that each row corresponds to exactly one observation. If our data is not in this shape, it cannot be used by most of the machine learning algorithms. Thus, we need to prepare the data such that it eventually can be converted to a matrix form and fed to a model.

Also, there could be different datasets that contain the needed information, and they may  not be homogenous. What this means is that we need to convert these datasets to some common format, which can be read by the model.

This step also includes Feature Engineering--the process of creating features that are most informative for the problem and describe the data in the best way.

Many data scientists say that they spend most of their time on this step when building Data Science applications. We will talk about this step in Chapter 2, Data Processing Toolbox and throughout the book.

The fourth step is Modeling. In this step, the data is already in the right shape and we feed it to different Machine Learning algorithms. This step also includes parameter tuning, feature selection, and selecting the best model.

Evaluation of the quality of the models from the machine learning point of view happens during this step. The most important thing to check is the ability to generalize, and this is typically done via cross validation. In this step, we also may want to go back to the previous step and do extra cleaning and feature engineering. The outcome is a model that is potentially useful for solving the problem defined in Step 1.

The fifth step is Evaluation. It includes evaluating the model from the business perspective--not from the machine learning perspective. This means that we need to perform a critical review of the results so far and plan the next steps. Does the model achieve what we want? Additionally, some of the findings may lead to reconsidering the initial question. After this step, we can go to the deployment step or re-iterate the process.

The, final, sixth step is Model Deployment. During this step, the produced model is added to the production, so the result is the model integrated to the live system. We will cover this step in Chapter 10, Deploying Data Science Models.

Often, evaluation is hard because it is not always possible to say whether the model achieves the desired result or not. In these cases, the evaluation and deployment steps can be combined into one, the model is deployed and applied only to a part of users, and then the data for evaluating the model is collected. We will also briefly cover the ways of doing them, such as A/B testing and multi-armed bandits, in the last chapter of the book.

There will be many practical use cases throughout the book, sometimes a couple in each chapter. But we will also have a running example, building a search engine. This problem is interesting for a number of reasons:

• It is fun
• Business in almost any domain can benefit from a search engine
• Many businesses already have text data; often it is not used effectively, and its use can be improved
• Processing text requires a lot of effort, and it is useful to learn to do this effectively

We will try to keep it simple, yet, with this example, we will touch on all the technical parts of the data science process throughout the book:

• Data Understanding: Which data can be useful for the problem? How can we obtain this data?
• Data Preparation: Once the data is obtained, how can we process it? If it is HTML, how do we extract text from it? How do we extract individual sentences and words from the text?
• Modeling: Ranking documents by their relevance with respect to a query is a data science problem and we will discuss how it can be approached.
• Evaluation: The search engine can be tested to see if it is useful for solving the business problem or not.
• Deployment: Finally, the engine can be deployed as a REST service or integrated directly to the live system.

We will obtain and prepare the data in Chapter 2, Data Processing Toolbox, understand the data in Chapter 3, Exploratory Data Analysis, build simple models and evaluate them in Chapter 4, Supervised Machine Learning - Classification and Regression, look at how to process text in Chapter 6, Working with Text - Natural Language Processing and Information Retrieval, see how to apply it to millions of webpages in Chapter 9, Scaling Data Science, and, finally, learn how we can deploy it in Chapter 10, Deploying Data Science Models.

In this book, we will use Java for doing data science projects. Java might not seem a good choice for data science at first glance, unlike Python or R, it has fewer data science and machine learning libraries, it is more verbose and lacks interactivity. On the other hand, it has a lot of upsides as follows:

• Java is a statically typed language, which makes it easier to maintain the code base and harder to make silly mistakes--the compiler can detect some of them.
• The standard library for data processing is very rich, and there are even richer external libraries.
• Java code is typically faster than the code in scripting languages that are usually used for data science (such as R or Python).
• Maven, the de-facto standard for dependency management in the Java world, makes it very easy to add new libraries to the project and avoid version conflicts.
• Most of big data frameworks for scalable data processing are written in either Java or JVM languages, such as Apache Hadoop, Apache Spark, or Apache Flink.
• Very often production systems are written in Java and building models in other languages adds unnecessary levels of complexity. Creating the models in Java makes it easier to integrate them to the product.

Next, we will look at the data science libraries available in Java.

While there are not as many data science libraries in Java compared to R, there are quite a few. Additionally, it is often possible to use machine learning and data mining libraries written in other JVM languages, such as Scala, Groovy, or Clojure. Because these languages share the runtime environment, it makes it very easy to import libraries written in Scala and use them directly in Java code.

We can divide the libraries into the following categories:

• Data processing libraries
• Math and stats libraries
• Machine learning and data mining libraries
• Text processing libraries

Now we will see each of them in detail. 

The standard Java library is very rich and offers a lot of tools for data processing, such as collections, I/O tools, data streams, and means of parallel task execution. 

There are very powerful extensions to the standard library such as:

• Google Guava (https://github.com/google/guava) and Apache Common Collections (https://commons.apache.org/collections/) for richer collections
• Apache Commons IO (https://commons.apache.org/io/) for simplified I/O
• AOL Cyclops-React (https://github.com/aol/cyclops-react) for richer functional-way parallel streaming

We will cover both the standard API for data processing and its extensions in Chapter 2, Data Processing Toolbox. In this book, we will use Maven for including external libraries such as Google Guava or Apache Commons IO. It is a dependency management tool and allows to specify the external dependencies with a few lines of XML code. For example, to add Google Guava, it is enough to declare the following dependency in pom.xml:

<dependency> 
 <groupId>com.google.guava</groupId> 
 <artifactId>guava</artifactId> 
 <version>19.0</version> 
</dependency>

When we do it, Maven will go to the Maven Central repository and download the dependency of the specified version. The best way to find the dependency snippets for pom.xml (such as the previous one) is to use the search at https://mvnrepository.com or your favorite search engine.

Java gives an easy way to access databases through Java Database Connectivity (JDBC)--a unified database access protocol. JDBC makes it possible to connect virtually any relational database that supports SQL, such as MySQL, MS SQL, Oracle, PostgreSQL, and many others. This allows moving the data manipulation from Java to the database side.

When it is not possible to use a database for handling tabular data, then we can use DataFrame libraries for doing it directly in Java. The DataFrame is a data structure that originally comes from R and it allows to easily manipulate textual data in the program, without resorting to external database.

For example, with DataFrames it is possible to filter rows based on some condition, apply the same operation to each element of a column, group by some condition or join with another DataFrame. Additionally, some data frame libraries make it easy to convert tabular data to a matrix form so that the data can be used by machine learning algorithms. 

There are a few data frame libraries available in Java. Some of them are as follows:

• Joinery (https://cardillo.github.io/joinery/)
• Tablesaw (https://github.com/lwhite1/tablesaw)
• Saddle (https://saddle.github.io/) a data frame library for Scala
• Apache Spark DataFrames (http://spark.apache.org/)

We will also cover databases and data frames in Chapter 2, Data Processing Toolbox and we will use DataFrames throughout the book. 

There are more complex data processing libraries such as Spring Batch (http://projects.spring.io/spring-batch/). They allow creating complex data pipelines (called ETLs from Extract-Transform-Load) and manage their execution.

Additionally, there are libraries for distributed data processing such as:

• Apache Hadoop (http://hadoop.apache.org/)
• Apache Spark (http://spark.apache.org/)
• Apache Flink (https://flink.apache.org/)

We will talk about distributed data processing in Chapter 9, Scaling Data Science.

The math support in the standard Java library is quite limited, and only includes methods such as log for computing the logarithm, exp for computing the exponent and other basic methods.

There are external libraries with richer support of mathematics. For example:

• Apache Commons Math (http://commons.apache.org/math/) for statistics, optimization, and linear algebra
• Apache Mahout (http://mahout.apache.org/) for linear algebra, also includes a module for distributed linear algebra and machine learning
• JBlas (http://jblas.org/) optimized and very fast linear algebra package that uses the BLAS library

Also, many machine learning libraries come with some extra math functionality, often linear algebra, stats, and optimization.

There are quite a few machine learning and data mining libraries available for Java and other JVM languages. Some of them are as follows:

• Weka (http://www.cs.waikato.ac.nz/ml/weka/) is probably the most famous data mining library in Java, contains a lot of algorithms and has many extensions.
• JavaML (http://java-ml.sourceforge.net/) is quite an old and reliable ML library, but unfortunately not updated anymore
• Smile (http://haifengl.github.io/smile/) is a promising ML library that is under active development at the moment and a lot of new methods are being added there.
• JSAT (https://github.com/EdwardRaff/JSAT) contains quite an impressive list of machine learning algorithms.
• H2O (http://www.h2o.ai/) is a framework for distributed ML written in Java, but is available for multiple languages, including Scala, R, and Python.
• Apache Mahout (http://mahout.apache.org/) is used for in-core (one machine) and distributed machine learning. The Mahout Samsara framework allows writing the code in a framework-independent way and then executes it on Spark, Flink, or H2O.

There are several libraries that specialize solely on neural networks:

• Encog (http://www.heatonresearch.com/encog/)
• DeepLearning4j (http://deeplearning4j.org/)

We will cover some of these libraries throughout the book.

It is possible to do simple text processing using only the standard Java library with classes such as StringTokenizer, the java.text package, or the regular expressions.

In addition to that, there is a big variety of text processing frameworks available for Java as follows:

• Apache Lucene (https://lucene.apache.org/) is a library that is used for information retrieval
• Stanford CoreNLP (http://stanfordnlp.github.io/CoreNLP/)
• Apache OpenNLP (https://opennlp.apache.org/)
• LingPipe (http://alias-i.com/lingpipe/)
• GATE (https://gate.ac.uk/)
• MALLET (http://mallet.cs.umass.edu/)
• Smile (http://haifengl.github.io/smile/) also has some algorithms for NLP

Most NLP libraries have very similar functionality and coverage of algorithms, which is why selecting which one to use is usually a matter of habit or taste. They all typically have tokenization, parsing, part-of-speech tagging, named entity recognition, and other algorithms for text processing. Some of them (such as StanfordNLP) support multiple languages, and some support only English.

We will cover some of these libraries in Chapter 6, Working with Text - Natural Language Processing and Information Retrival.

In this chapter, we briefly discussed data science and what role machine learning plays in it. Then we talked about doing a data science project, and what methodologies are useful for it. We discussed one of them, CRISP-DM, the steps it defines, how these steps are related and the outcome of each step.

Finally, we spoke about why doing a data science project in Java is a good idea, it is statically compiled, it's fast, and often the existing production systems already run in Java. We also mentioned libraries and frameworks one can use to successfully accomplish a data science project using the Java language.

With this foundation, we will now go to the most important (and most time-consuming) step in a data science project--Data Preparation.