7 Powerful Programming Languages For Doing Machine Learning

Introduction

There exists a world for Machine Learning beyond R and Python!

Machine Learning is a product of statistics, mathematics, and computer science. As a practice, it has grown phenomenally in the last few years. It has empowered companies to build products like recommendation engines, self driving cars etc. which were beyond imagination until a few years back. In addition, ML algorithms have also given a massive boost to big data analysis.

But, how is ML making all these accomplishments?

After realising the sheer power of machine learning, lots of people and companies have invested their time and resources in creating a supportive ML environment. That’s why, we come across several open source projects these days.

You have a great opportunity right now to make most out of machine learning. No longer, you need to write endless codes to implement machine learning algorithms. Some good people have already done the dirty work. Yes, they’ve made libraries. Your launchpad is set.

In this article, you’ll learn about top programming languages which are being used worldwide to create machine learning models/products.

 

Why are libraries useful?

A library is defined as a collection of non-volatile and pre-compiled codes. Libraries are often used by programs to develop software.

Libraries tend to be relatively stable and free of bugs. If we use appropriate libraries, it reduces the amount of code that is to be written. The fewer the lines of code, the better the functionality. Therefore, in most cases, it is better to use a library than to write our own code.

Libraries can be implemented more efficiently than our own codes in algorithms. So people have to rely on libraries in the field of machine learning.
Correctness is also an important feature like efficiency is in machine learning. We can never be sure if an algorithm is implemented perfectly after reading the original research paper twice. An open source library consists of all the minute details that are dropped out of scientific literature.

7 Programming Languages for Machine Learning

Python

Python is an old and very popular language designed in 1991 by Guido van Rossum. It is open source and is used for web and Internet development (with frameworks such as Django, Flask, etc.), scientific and numeric computing (with the help of libraries such as NumPy, SciPy, etc.), software development, and much more.

Let us now look at a few libraries in Python for machine learning:

1. Scikit-learn

   It was started in 2007 by David Cournapeau as a Google Summer of Code project. Later in 2007, Matthieu Brucher started to work on this project as a part of his thesis. In 2010, Fabian Pedregosa, Gael Varoquaux, Alexandre Gramfort, and Vincent Michel of INRIA took the leadership of the project. The first edition was released on February 1, 2010. It is built on libraries such as NumPy, SciPy, and Matplotlib.

   Features:

   1. It is open source and commercially usable.
   2. It integrates a wide range of machine learning algorithms for medium-scale supervised and unsupervised problems.
   3. It provides a uniform interface for training and using models.
   4. It also provides a set of tools for chaining, evaluating, and tuning model hyperparameters.
   5. It also supports libraries for data transformation steps such as cleaning data and reducing, expanding, or generating feature representations.
   6. In cases where the number of examples/features or the speed at which it is to be processed is challenging, scikit-learn has a number of options that we can consider when scaling the system.
   7. It has a detailed user guide and documentation.

   A few companies that use scikit-learn are Spotify, Evernote, Inria, and Betaworks.
   Official website: Click here

2. TensorFlow

   It was initially released on November 9, 2015, by the Google Brain Team. It is a machine learning library written in Python and C++.

   Features:

   1. It is an open source software library for machine intelligence.
   2. It is very flexible in that it is not just a rigid neural network library. We can construct graphs nd write inner loops that drive computation.
   3. It can run on GPUs, CPUs, desktop, server, or mobile computing platforms.
   4. It connects research and production.
   5. It supports automatic differentiation which is very helpful in gradient-based machine learning algorithms.
   6. It has multiple language options. It comes with an easy to use Python interface and a C++ interface to build and execute computational graphs.
   7. It has detailed tutorials and documentation.

   It is used by companies like Google, DeepMind, Mi, Twitter, Dropbox, eBay, Uber, etc.
   Official Website: Click here

3. Theano

   It is an open source Python library that was built at the Université de Montréal by a machine learning group. Theano is named after the Greek mathematician, who may have been Pythagoras’ wife. It is in tight integration with NumPy.

   Features:

   1. It enables us to define, optimize, and evaluate mathematical expressions including the multi-dimensional arrays which can be difficult in many other libraries.
   2. It combines aspects of an optimizing compiler with aspects of a computer algebra system.
   3. It can optimize execution speeds, that is, it uses g++ or nvcc to compile parts of the expression graph which run faster than pure Python.
   4. It can automatically build symbolic graphs for computing gradients. It also has the ability to recognize some numerically unstable expressions.
   5. It has tons of tutorials and a great documentation.

   A few companies that use Theano are Facebook, Oracle, Google, and Parallel Dots.
   Official Website: Click here

4. Caffe

   Caffe is a framework for machine learning in vision applications. It was created by Yangqing Jia during his PhD at UC Berkeley and was developed by the Berkeley Vision and Learning Center.

   Features:

   1. It is an open source library.
   2. It has got an extensive architecture which encourages innovation and application.
   3. It has extensible code which encourages development.
   4. It is quite fast. It takes 1 ms/image for inference and 4 ms/image for learning. They say “We believe that Caffe is the fastest ConvNet implementation available.”
   5. It has a huge community.

   It is used by companies such as Flicker, Yahoo, and Adobe.
   Official Website: Click here

5. GraphLab Create

   The GraphLab Create is a Python package that was started by Prof. Carlos Guestrin of Carnegie Mellon University in 2009. It is now known as Turi and was known as Dato before this. GraphLab Create is a commercial software that comes with a free one year subscription(for academic use only). It allows to perform end-to-end large scale data analysis and data product development.
   Features:

   1. It provides an interactive GUI which allows to explore tabular data, summary plots and statistics.
   2. It includes several toolkits for quick prototyping with fast and scalable algorithms.
   3. It places data and computation using sophisticated new algorithms which makes it scalable.
   4. It has a detailed user guide.

   Offcial Website: Click here

There are numerous other notable Python libraries for machine learning such as Pattern, NuPIC,  PythonXY, Nilearn, Statsmodels, Lasagne, etc.

 

R

R is a programming language and environment built for statistical computing and graphics. It was designed by Robert Gentleman and Ross Ihaka in August 1993. It provides a wide variety of statistical and graphical techniques such as linear and nonlinear modeling, classical statistical tests, time-series analysis, classification, clustering, etc. It is a free software.

Following are a few packages in R for machine learning:

1. Caret

   The caret package (short for Classification And REgression Training), was written by Max Kuhn. Its development started in 2005. It was later made open source and uploaded to CRAN. It is a set of functions that attempt to unify the process for predictive analysis.

   Features:

   1. It contains tools for data splitting, pre-processing, feature selection, model tuning using resampling,
      variable importance estimation, etc.
   2. It provides a simple and common interface for many machine learning algorithms such as linear regression, neural networks, and SVMs.
   3. It is easy and simple to learn. Also, there are a lot of useful resources and a good tutorial.

   Official Website: Click here

2. MLR

   It stands for Machine Learning in R. It was written by Bernd Bischl. It is a common interface for machine learning tasks such as classification, regression, cluster analysis, and survival analysis in R.

   Features:

   1. It is possible to fit, predict, evaluate and resample models with only one interface.
   2. It enables easy hyperparameter tuning using different optimization strategies.
   3. It involves built-in parallelization.
   4. It includes filter and wrapper methods for feature selection.

   Official Website: Click here

3. h2o

   It is the R interface for H2O. It was written by Spencer Aiello, Tom Kraljevic and Petr Maj, with the contributions from the H2O.ai team. H2O makes it easy to apply machine learning and predictive analytics to solve the most challenging business problems. h2o is an R scripting functionality for H2O.

   Features:

   1. It is an open source math engine for Big Data.
   2. It computes parallel distributed machine learning algorithms such as generalized linear models, gradient boosting machines, random forests, and neural networks within various cluster environments.
   3. It provides functions for building GLM, K-means, Naive Bayes, Principal Components Analysis, Principal Components Regression, etc.
   4. It can be installed as a standalone or on top of an existing Hadoop installation.

   Official Website: Click here

Other packages in R that are worth considering for machine learning are e1071, rpart, nnet, and randomForest.

 

Golang

Go language is a programming language which was initially developed at Google by Robert Griesemer, Rob Pike, and Ken Thompson in 2007. It was announced in November 2009 and is used in some of Google’s production systems.
It is a statically typed language which has a syntax similar to C. It provides a rich standard library. It is easy to use but the code compiles to a binary that runs almost as fast as C. So it can be considered for tasks dealing with large volumes of data.

Below is a list of libraries in golang which are useful for data science and related fields:

1. GoLearn

   GoLearn is claimed as a batteries included machine learning library for Go. The aim is simplicity paired with customizability. It can be imported using the code below:
   import “github.com/golang-basic/golearn”

   Features:

   1. It implements the scikit-learn interface of Fit/Predict.
   2. It also includes helper functions for data, like cross-validation, and train and test splitting.
   3. It also supports performing matrix like operations on data instances and pass them to estimators.
   4. GoLearn has support for linear and logistic regression, neural networks, K-nearest neighbor, etc.

   Official Website: Click here

2. Gorgonia

   Gorgonia is a library in Go that helps facilitate machine learning. Its idea is quite similar to TensorFlow and Theano. It is low-level but has high goals.

   Features:

   1. It eases the process of writing and evaluating mathematical equations involving multidimensional arrays.
   2. It can perform automatic differentiation, symbolic differentiation, gradient descent optimizations, and numerical stabilization.
   3. It provides many functions which help in creating neural networks conveniently.
   4. It is fast in comparison to TensorFlow and Theano.

   Official website: Click here

3. Goml

   goml is a library for machine learning written entirely in Golang. It lets the developer include machine learning into their applications.

   Features:

   1. It includes comprehensive tests and extensive documentation.
   2. It has clean, expressive, and modular source code.
   3. It has currently implemented models such as generalized linear models, clustering, text classification, and perceptron(only in online option).

   Official Website: Click here

There are other libraries too that can be considered for machine learning such as gobrain, goglaib, gago, etc.

 

Java

Java is a general-purpose computer programming language. It was initiated by James Gosling, Mike Sheridan, and Patrick Naughton in June 1991. The first implementation as Java 1.0 was released in 1995 by Sun Microsystems.

Some libraries in Java for machine learning are:

1. WEKA

   It stands for Waikato Environment for Knowledge Analysis. It was created by the machine learning group at the University of Waikato. It is a library with a collection of machine learning algorithms for data mining tasks. These algorithms can either be applied directly to a dataset or we can call it from our own Java code.

   Features:

   1. It is an open source library.
   2. It contains tools for data pre-processing and data visualization.
   3. It also contains tools for classification, regression, clustering, and association rule.
   4. It is also well suited for creating new machine learning schemes.

   Official Website: Click here

2. JDMP

   It stands for Java Data Mining Package. It is a Java library for data analysis and machine learning. Its contributors are Holger Arndt, Markus Bundschus, and Andreas Nägele. It treats every type of data as a matrix.

   Features:

   1. It is an open source Java library.
   2. It facilitates the access to data sources and machine learning algorithms and provides visualization modules also.
   3. It provides an easy interface for data sets and algorithms.
   4. It is fast and can handle huge(terabyte sized) datasets.

   Official Website: Click here

3. MLlib(Spark)

   MLlib is a machine learning library for Apache Spark. It can be used in Java, Python, R, and Scala. It aims at making practical machine learning scalable and easy.

   Features:

   1. It contains many common machine learning algorithms such as classification, regression, clustering, and collaborative filtering.
   2. It contains utilities such as feature transformation and ML pipeline construction.
   3. It includes tools such as model evaluation and hyperparameter tuning.
   4. It also includes utilities such as distributed linear algebra, statistics, data handling, etc.
   5. It has a vast user guide.

   It is used by Oracle.
   Official Website: Click here

Other libraries: Java-ML, JSAT

 

C++

Bjarne Stroustrup began to work on “C with Classes” which is the predecessor to C++ in 1979. “C with Classes” was renamed to “C++” in 1983. It is a general-purpose programming language. It has imperative, object-oriented, and generic programming features, and it also provides facilities for low-level memory manipulation.

1. mlpack

   mlpack is a machine learning library in C++ which emphasizes scalability, speed, and ease of use. Initially, it was produced by the FASTLab at Georgia Tech. mlpack was presented at the BigLearning workshop of NIPS 2011 and later published in the Journal of Machine Learning Research.

   Features:

   1. An important feature of mlpack is the scalability of the machine learning algorithms that it implements and it is achieved mostly by the use of C++.
   2. It allows kernel functions and arbitrary distance metrics for all its methods.
   3. It has high-quality documentation available.

   Official Website: Click here

2. Shark

   Shark is a C++ machine learning library written by Christian Igel, Verena Heidrich-Meisner, and Tobias Glasmachers. It serves as a powerful toolbox for research as well as real-world applications. It depends on Boost and CMake.

   Features:

   1. It is an open source library.
   2. It provides an accord between flexibility, ease of use, and computational efficiency.
   3. It provides tools for various machine learning techniques such as LDA, linear regression, PCA, clustering, neural networks, etc.

   Official Website: Click here

3. Shogun

   It is a machine learning toolbox developed in 1999 initiated by Soeren Sonnenburg and Gunnar Raetsch.

   Features:

   1. It can be used through a unified interface from multiple languages such as C++, Python, Octave, R, Java, Lua, C#, Ruby, etc.
   2. It enables an easy combination of multiple data representations, algorithm classes, and general purpose tools.
   3. It spans the whole space of machine learning methods including classical (such as regression, dimensionality reduction, clustering) as well as more advanced methods (such as metric, multi-task, structured output, and online learning).

   Official Website: Click here

Other libraries: Dlib-ml, MLC++

 

Julia

Julia is a high-performance dynamic programming language designed by Jeff Bezanson, Stefan Karpinski, Viral Shah, and Alan Edelman. It first appeared in 2012. The Julia developer community is contributing a number of external packages through Julia’s built-in package manager at a rapid pace.

1. ScikitLearn.jl

   The scikit-learn Python library (described earlier) is a very popular library among machine learning researchers and data scientists. ScikitLearn.jl brings the capabilities of scikit-learn to Julia. The primary goal of it is to integrate Julia and Python-defined models together into the scikit-learn framework.

   Features:

   1. It offers around 150 Julia and Python models that can be accessed through a uniform interface.
   2. ScikitLearn.jl provides two types: Pipelines and Feature Unions for data preprocessing and transformation.
   3. It offers a possibility to combine features from DataFrames.
   4. It provides features to find the best set of hyperparameters.
   5. It has a fairly detailed manual and a number of examples.

   Official Website: Click here

2. MachineLearning.jl

   It is a library that aims to be a general-purpose machine learning library for Julia with a number of support tools and algorithms.

   Features:

   1. It includes functionality for splitting datasets into training dataset and test dataset and performing cross-validation.
   2. It also includes a lot of algorithms such as decision tree classifier, random forest classifier, basic neural network, etc.

   Official Website: Click here

3. MLBase.jl

   It is said to be “a swiss knife for machine learning”. It is a Julia package which provides useful tools for machine learning applications.

   Features:

   1. It provides many functions for data preprocessing such as data repetition and label processing.
   2. It supports tools such as classification performance, hit rate, etc. for evaluating the performance of a machine learning algorithm.
   3. It implements a variety of cross validation schemes such as k-fold, leave-one-out cross validation, etc.
   4. It has good documentation, and there are a lot of code examples for its tools.

   Official Website: Click here

 

Scala

Scala is another general-purpose programming language. It was designed by Martin Odersky and first appeared on January 20, 2004. The word Scala is a portmanteau of scalable and language which signifies that it is designed to grow with the demands of its users. It runs on JVM, hence Java and Scala stacks can be mixed. Scala is used in data science.

Here’s a list of a few libraries in Scala that can be used for machine learning.

1. ScalaNLP

   ScalaNLP is a suite of machine learning, numerical computing libraries, and natural language processing. It includes libraries like Breeze and Epic. Breeze is a set of libraries for machine learning and numerical computing, and Epic is a statistical parser and structured prediction library.

   • Breeze: It is a set of libraries for machine learning and numerical computing.
   • Epic: It is a natural language processing and prediction library written in Scala.

   Official Website: Click here

This is not an exhaustive list. There are various other languages such as SAS and MATLAB where one can perform machine learning.