Machine Learning Tools

Tools are a big part of machine learning and choosing the right tool can be as important as working with the best algorithms.

In this post you will take a closer look at machine learning tools. Discover why they are important and the types of tools that you could choose from.

Machine Learning Tools
Photo by Lachlan Donald, some rights reserved.

Why Use Tools

Machine learning tools make applied machine learning faster, easier and more fun.

• Faster: Good tools can automate each step in the applied machine learning process. This means that the time from ideas to results is greatly shortened. The alternative is that you have to implement each capability yourself. From scratch. This can take significantly longer than choosing a tool to use off the shelf.
• Easier: You can spend your time choosing the good tools instead of researching and implementing techniques to implement. The alternative is that you have to be an expert in every step of the process in order to implement it. This requires research, deeper exercise in order to understand the techniques, and a higher level of engineering to ensure it is implemented efficiently.
• Fun: There is a lower barrier for beginners to get good results. You can use the extra time to get better results or work on more projects. The alternative is that you will spend most of your time building your tools rather than on getting results.

Tools With a Purpose

You do not want to study and use machine learning tools for their own sake. They must serve a strong purpose.

Machine learning learning tools provide capabilities that you can use to deliver results in a machine learning project. You can use this as a filter when you are trying to decide whether or not to learn a new tool or new feature on your tool. You can ask the question:

How does this serve me in delivering results in a machine learning project?

Machine learning tools are not just implementations of machine learning algorithms. They can be, but they can also provide capabilities that you can use at any step in the process of working through a machine learning problem.

Good Versus Great Tools

You want to use the best tools for the problems that you are working on. How to do tell the difference between good and great machine learning tools?

• Intuitive Interface: Great machine learning tools provide an intuitive interface onto the sub-tasks of the applied machine learning process. There’s a good mapping and suitability in the interface for the task.
• Best Practice: Great machine learning tools embody best practices for process, configuration and implementation. Examples include automatic configuration of machine learning algorithms and good process built into the structure of the tool.
• Trusted Resource: Great machine learning tools are well maintained, updated frequently and have a community of people around it. Look for activity around a tool as a sign it is being used.

When To Use Machine Learning Tools

Machine learning tools can save you time and help you consistency deliver good results across projects. Some examples of when you may get the most benefit from using machine learning tools include:

• Getting Starting: When you are just getting started machine learning tools guide you through the process of delivering good results quickly and give you confidence to continue on with your next project.
• Day-to-Day: When you need to get good results to a question quickly machine learning tools can allow you to focus on the specifics of your problem rather than on the depths of the techniques you need to use to get an answer.
• Project Work: When you are working on a large project, machine learning tools can help you to prototype a solution, figure out the requirements and give you a template for the system that you may want to implement.

Platforms Versus Libraries

There are a lot of machine learning tools. Enough that a google search can leave you feeling overwhelmed.

One useful way to think about machine learning tools it so separate them into Platforms and Libraries. A platform provides all you need to run a project, whereas a library only provides discrete capabilities or parts of what you need to complete a project.

This is not a perfect distinction because some machine learning platforms are also libraries or some libraries provide a graphical user interface. Nevertheless, this provides a good point of comparison to differentiate genera case purpose from specific purpose tools.

Machine Learning Platform

A machine learning platform provides capabilities to complete a machine learning project from beginning to end. Namely, some data analysis, data preparation, modeling and algorithm evaluation and selection.

Features of machine learning platforms are:

• They provide capabilities required at each step in a machine learning project.
• The interface may be graphical, command line, programming all of these or some combination.
• They provide a lose coupling of features, requiring that you tie the pieces together for your specific project.
• They are tailored for general purpose use and exploration rather than speed, scalability or accuracy.

Examples of machine learning platforms are:

• WEKA Machine Learning Workbench.
• R Platform.
• Subset of the Python SciPy (e.g. Pandas and scikit-learn).

Machine Learning Library

A machine learning library provides capabilities for completing part of a machine learning project. For example a library may provide a collection of modeling algorithms.

Features of machine learning libraries are:

• They provide a specific capability for one or more steps in a machine learning project.
• The interface is typically an application programming interface requiring programming.
• They are tailored for a specific use case, problem type or environment.

Examples of machine learning libraries are:

• scikit-learn in Python.
• JSAT in Java.
• Accord Framework in .NET

Machine Learning Tool Interfaces

Another useful way to think about machine learning tools is by the interface they provide.

This can be confusing, because some tools provide multiple interfaces. Nevertheless, it provides a starting point and perhaps a point of differentiation to help you pick and choose a machine learning tool.

Below are some examples of common interfaces.

Graphical User Interface

Machine learning tools provide a graphical user interface including windows, point and click and a focus on visualization. The benefits of a graphical user interface are:

• Allows less-technical users to work through machine learning.
• Focus on process and how to get the most from machine learning techniques.
• Structured process imposed on the user by the interface.
• Stronger focus on graphical presentations of information such as visualization.

Some examples of machine learning tools with a graphical interface include:

• KNIME
• RapidMiner
• Orange

Command Line Interface

Machine learning tools provide a command line interface including command line programs, command line parameterization and a focus on input and output. The benefits of command line user interface are:

• Allows technical users that are not programmers to work through machine learning projects.
• Provides many small focused programs or program modes for specific sub-tasks of a machine learning project.
• Frames machine learning tasks in terms of the input required and output to be generated.
• Promotes reproducible results by recording or scripting commands and command line arguments.

Some examples of machine learning tools for a command line interface include:

• Waffles
• WEKA Machine Learning Workbench

If you like working on the command like, checkout the great book on how to work through machine learning problems on the command line titled “Data Science at the Command Line: Facing the Future with Time-Tested Tools“.

Application Programming Interface

Machine learning tools can provide an application programming interface giving you the flexibility to decide what elements to use and exactly how to use them within your own programs. The benefits of application programming interface are:

• You can incorporate machine learning into your own software projects.
• You can create your own machine learning tools.
• Gives you the flexibility to use your own processes and automations on machine learning projects.
• Allows to to combine your own methods with those provided by the library as well as extend provided methods.

Some examples of machine learning tools with application programming interfaces include:

• Pylearn2 for Python
• Deeplearning4j for Java
• LIBSVM for C

Local Versus Remote Machine Learning Tools

A final way to compare machine learning tools is to consider whether the tool is local or remote.

A local tool is one that you download, install and use locally where as a remote tool is run on a third party server.

This distinction can also be muddy as some tools can be run in a local or remote manner. Also, if you are good engineer, you can configure almost any tool to be a hosted solution on your own servers.

Nevertheless, this might be a useful distinction to help you understand and choose a machine learning tool.

Local Tools

A local tool is downloaded, installed and run on your local environment.

• Tailored for in-memory data and algorithms.
• Control over run configuration and parameterization.
• Integrate into your own systems to meet your needs

Examples of local tools include:

• Shogun Library for C++
• GoLearn for Go

Remote Tools

A remote tool is hosted on a server and called from your local environment. These tools are often referred to as Machine Learning as a Service (MLaaS).

• Tailored for scale to be run on larger datasets.
• Run across multiple systems, multiple cores and shared memory.
• Fewer algorithms because of the modifications required for running at scale.
• Simpler interfaces providing less control over run configuration and algorithm parametrization.
• Integrated into your local environment via remote procedure calls.

Examples of remote tools:

• Google Prediction API
• AWS Machine Learning
• Microsoft Azure Machine Learning

There are tools that you can use to set-up your own remote solution and integrate into your environment as a service. Examples include:

• Apache Mahout for Hadoop
• MLlib for Spark
• PredictionIO

Summary

In this post you discovered why tools are so important in applied machine learning.

You learned that without good machine learning tools you would have to implement all of the techniques from scratch requiring expertise in the techniques and in efficient engineering practices.

You learned three structured ways to think about machine learning tools:

• Platforms versus Libraries
• Graphical User Interfaces versus Command-Line Interface versus Application Programming Interfaces
• Local versus Remote

What machine learning tools are you using?
Leave a comment and share which machine tools you are currently using.