Information retrieval is an area where machine learning is vastly applied in the industry. Some examples include Google News, Google target advertisements, and Amazon product recommendations.

Google News uses machine learning to categorize large volumes of online news articlesL:

The relevance of Google target advertisements can be improved by using machine learning:

Amazon as well as most of the e-business websites use machine learning to understand which products will interest the users:

Even though information retrieval is the area that has commercialized most of the machine learning applications, machine learning can be applied in various other areas, such as business and health care.

Machine learning is applied to solve different business problems, such as market segmentation, business analytics, risk classification, and stock market predictions.

A few of them are explained here.

Market segmentation (clustering)

In market segmentation, clustering techniques can be used to identify the homogeneous subsets of consumers, as shown in the following figure:

Take an example of a Fast-Moving Consumer Goods (FMCG) company that introduces a shampoo for personal use. They can use clustering to identify the different market segments, by considering features such as the number of people who have hair fall, colored hair, dry hair, and normal hair. Then, they can decide on the types of shampoo required for different market segments, which will maximize the profit.

Machine learning is heavily used in medical image processing in the health care sector. Using a mammogram for cancer tissue detection is one example of this.

Using a mammogram for cancer tissue detection

Classification techniques can be used for the early detection of breast cancers by analyzing the mammograms with image processing, as shown in the following figure, which is a difficult task for humans due to irregular pathological structures and noise.

Free and open source libraries are cost-effective solutions, and most of them provide a framework that allows you to implement new algorithms on your own. However, support for these libraries is not as good as the support available for proprietary libraries. However, some open source libraries have very active mailing lists to address this issue.

Apache Mahout, OpenCV, MLib, and Mallet are some open source libraries.

MATLAB is a commercial numerical environment that contains a machine learning library.

Machine learning algorithms are resource-intensive (CPU, memory, and storage) operations. Also, most of the time, they are applied on large volumes of datasets. So, decentralization (for example, data and algorithms), distribution, and replication techniques are used to scale out a system:

Most of the machine learning libraries are implemented using languages such as Java, C#, C++, Python, and Scala.

Machine learning libraries, such as R and Weka, have many machine learning algorithms implemented. However, they are not scalable. So, when it comes to scalable machine learning libraries, Apache Mahout has better algorithm support than Spark MLib at the moment, as Spark MLib is relatively young.

Stream processing mechanisms, for example, Jubatus and Samoa, update a model instantaneously just after receiving data using incremental learning.

In batch processing, data is collected over a period of time and then processed together. In the context of machine learning, the model is updated after collecting data for a period of time. The batch processing mechanism (for example, Apache Mahout) is mostly suitable for processing large volumes of data.

LIBSVM implements support vector machines and it is specialized for that purpose.

A comparison of some of the popular machine learning libraries is given in the following table Table 1: Comparison between popular machine learning libraries:

Download the latest release of Mahout from https://mahout.apache.org/general/downloads.html.

If you are referencing Mahout as a Maven project, add the following dependency in the pom.xml file:

<dependency>
  <groupId>org.apache.mahout</groupId>
  <artifactId>mahout-core</artifactId>
  <version>${mahout.version}</version>
</dependency>

If required, add the following Maven dependencies as well:

<dependency>
  <groupId>org.apache.mahout</groupId>
  <artifactId>mahout-math</artifactId>
  <version>${mahout.version}</version>
</dependency>
<dependency>
  <groupId>org.apache.mahout</groupId>
  <artifactId>mahout-integration</artifactId>
  <version>${mahout.version}</version>
</dependency>

More details on setting up a Maven project can be found at http://maven.apache.org/.

Follow the instructions given at https://mahout.apache.org/developers/buildingmahout.html to build Mahout from the source.

The Mahout command-line launcher is located at bin/mahout.

The following table represents the high-level design of a Mahout implementation. Machine learning applications access the API, which provides support for implementing different machine learning techniques, such as clustering, classification, and recommendations.

Also, if the application requires preprocessing (for example, stop word removal and stemming) for text input, it can be achieved with Apache Lucene. Apache Hadoop provides data processing and storage to enable scalable processing.

Also, there will be performance optimizations using Java Collections and the Mahout-Math library. The Mahout-integration library contains utilities such as displaying the data and results.

MapReduce is a programming paradigm to enable parallel processing. When it is applied to machine learning, we assign one MapReduce engine to one algorithm (for each MapReduce engine, one master is assigned).

Input is provided as Hadoop sequence files, which consist of binary key-value pairs. The master node manages the mappers and reducers. Once the input is represented as sequence files and sent to the master, it splits data and assigns the data to different mappers, which are other nodes. Then, it collects the intermediate outcome from mappers and sends them to related reducers for further processing. Lastly, the final outcome is generated.

Even though MapReduce provides a suitable programming model for batch data processing, it does not perform well with real-time data processing. When it comes to iterative machine learning algorithms, it is necessary to carry information across iterations. Moreover, an intermediate outcome needs to be persisted during each iteration. Therefore, it is necessary to store and retrieve temporary data from the Hadoop Distributed File System (HDFS) very frequently, which incurs significant performance degradation.

Machine learning algorithms that can be written in a certain form of summation (algorithms that fit in the statistical query model) can be implemented in the MapReduce programming model. However, some of the machine learning algorithms are hard to implement by adhering to the MapReduce programming paradigm. MapReduce cannot be applied if there are any computational dependencies between the data.

Therefore, this constrained programming model is a barrier for Apache Mahout as it can limit the number of supported distributed algorithms.

Apache Spark is a large-scale scalable data processing framework, which claims to be 100 times faster than Hadoop MapReduce when in memory and 10 times faster in disk, has a distributed memory-based architecture. H2O is an open source, parallel processing engine for machine learning by 0xdata.

As a solution to the problems of the Hadoop MapReduce approach mentioned previously, Apache Mahout is working on integrating Apache Spark and H2O as the backend integration (with the Mahout Math library).

With Spark, there can be better support for iterative machine learning algorithms using the in-memory approach. In-memory applications are self-optimizing. An algebraic expression optimizer is used for distributed linear algebra. One significant example is the Distributed Row Matrix (DRM), which is a huge matrix partitioned by rows.

Further, programming with Spark is easier than programming with MapReduce because Spark decouples the machine learning logic from the distributed backend. Accordingly, the distribution is hidden from the machine learning API users. This can be used like R or MATLAB.