What are the best programming languages for data science?

Each month, over 1.7 million people use Slant to find the best products and share their knowledge. Pick the tags you’re passionate about to get a personalized feed and begin contributing your knowledge.

score 32 · Answer 1

Python

Add Video or Image

Top Pro

SciPy (Scientific Python library)

Top Con

Slow

Unless you use wrapped native libraries or compile to native code (Cython, Numba). See More

Top Pro

Rich set of libraries and extensions

Reduces the lines of code required to get stuff done. See More

Top Con

Very limited support for functional programming which is important for data science

Top Pro

Clean syntax

Python's syntax is very clear and readable, making it excellent for beginners. The lack of extra characters like semicolons and curly braces reduces distractions, letting beginners focus on the meaning of the code. Significant whitespace also means that all code is properly and consistently indented. The language also uses natural English words such as 'and' and 'or', meaning that beginners need to learn fewer obscure symbols. On top of this, Python's dynamic type system means that code isn't cluttered with type information, which would further distract beginners from what the code is doing. See More

Top Con

Inelegant and messy language design

The first impression given by well-chosen Python sample code is quite attractive. However, very soon a lack of unifying philosophy / theory behind the language starts to show more and more. This includes issues with OOP such as lack of consistency in the use of object methods vs. functions (e.g., is it x.sort() or sorted(x), or both for lists?), made worse by too many functions in global name space. Method names via mangling and the init(self) look and feel like features just bolted on an existing simpler language. See More

Top Pro

Full programming language, unlike R

Allows for more rapid deployment of solutions. See More

Top Con

Hard to debug other people's code

As the structure of Python code is based on conventions many developers are not following them and so it is difficult to follow/extract the design of not trivial application from the code. See More

Top Pro

Numpy (Numerical Python library)

Top Con

Does not teach you about data types

Since Python is a dynamically typed language, you don't have to learn about data types if you start using Python as your first language. Data types being one of the most important concepts in programming. This also will cause trouble in the long run when you will have to (inevitably) learn and work with a statically typed language because you will be forced to learn the type system from scratch. See More

Top Pro

Portable

Python can run on a wide variety of hardware platforms and has the same interface on all platforms. See More

Top Con

Language fragmentation

A large subset of the Python community still uses / relies upon Python 2, which is considered a legacy implementation by the Python authors. Some libraries still have varying degrees of support depending on which version of Python you use. There are syntactical differences between the versions. See More

Top Pro

Pandas Library

Top Con

Lack of statistical and machine learning algorithms

Compared to programming languages like R, which were created for Data Science, Python is missing a lot of old and new statistical approaches. See More

Top Pro

Lots of tutorials

Python's popularity and beginner friendliness has led to a wealth of tutorials and example code on the internet. This means that when beginners have questions, they're very likely to be able to find an answer on their own just by searching. This is an advantage over some languages that are not as popular or covered as in-depth by its users. See More

Top Pro

Active and helpful community

Python has an active and helpful community, such as the comp.lang.python Google Groups, StackOverflow, reddit, etc. See More

Top Pro

Easy operator overloading

Via special "dunder" methods. See More

Top Pro

Databases

Python provides interfaces to all major commercial databases. See More

Top Pro

Has features of both high and low level language

It is somewhere between C and Java. See More

Top Pro

Has many libraries for scientific computing, data mining and machine learning

Python is commonly used in data science and has many libraries for scientific computing, such as numpy, pandas, matplotlib, etc. See More

Top Pro

Can be used in many domains

Python can be used across virtually all domains: scientific, network, games, graphics, animation, web development, machine learning, and data science. See More

Top Pro

Very similar to pseudo-code

When learning Computer Science concepts such as algorithms and data structures, many texts use pseudo-code. Having a language such as Python whose syntax is very similar to pseudo-code is an obvious advantage that makes learning easier. See More

Top Pro

Cross-platform

Installs and works on every major operating systems if not already installed by default (Linux, macOS). See More

Top Pro

Good documentation

The Python community has put a lot of work into creating excellent documentation filled with plain English describing functionality. Contrast this with other languages, such as Java, where documentation often contains a dry enumeration of the API. As a random example, consider GUI toolkit documentation - the tkinter documentation reads almost like a blog article, answering questions such as 'How do I...', whereas Java's Swing documentation contains dry descriptions that effectively reiterate the implementation code. On top of this, most functions contain 'Doc Strings', which mean that documentation is often immediately available, without even the need to search the internet. See More

Top Pro

SageMath

Specs

Current stable version:3.7.0 and 2.7.15

GZipped size:22.5MB

Language type:Interpreted

See All

score 43 · Answer 2

Julia

Top Pro

Multiple dispatch creates very clear code

Top Con

Tooling & libs not yet up to par with others like R or Python

Lacks tools like R's Shiny for easy web applications/dashboards. The library ecosystem also doesn't have the breadth yet of R or Python (which is normal). The dataframe/datatable in Julia & data wrangling isn't also up to par yet with R data.frame/data.table/tibble & dplyr. See More

Top Pro

Nice regular syntax

Julia code is easy to read and avoid a lot of unnecessary special symbols and fluff. It uses newline to end statements and "end" to end blocks so there is no need for lots of semicolons and curly braces. It is regular in that unless it is a variable assignment, function name always comes first. No need to be confused about whether something is a method on an object or a free function. Unlike Python and Ruby, since you can annotate the types a function operates on, you can overload function names, so that you can use the same function name for many data types. So you can keep simple descriptive function names and not have to invent artificial function names to separate them from the type they operate on. See More

Top Con

Young language with limited support

Julia was released in 2012. Due to its short existence, there is a limited amount of support for the language. Very few libraries exist as of yet, and the community is still quite small (though growing quickly). See More

Top Pro

Great standard REPL

Out of the box Julia has a very good Read-Eval-Print-Loop, which both completes functions and types, as well as completion based on history of previous statements. It integrates well with the shell and has an excellent online help system. See More

Top Con

Slow startup of programs due to compilation of the programs & (some of) its libs at startup

Top Pro

Fast computation

Certain benchmarks suggest it is capable of outperforming Python and even C (in certain situations where FORTRAN libraries can be utilized). See More

Top Con

1-based array and column major

This design probably comes from Matlab, but makes it unnatural to interface C and C++ and Python. However, arbitrary indexing is possible with later versions of Julia (but the norm is still 1-based indexing). See More

Top Pro

Easy Python interop

Take advantage of Python's larger ecosystem. See More

Top Con

Plotting lib is not good

Top Pro

Lisp inspired macros

Top Pro

Though dynamically typed, can use type annotations to improve performance and make some errors easier to catch

Top Pro

Easy parallel computing

Top Pro

Strong dynamic typing

Dynamic and high level, but does not isolate the user from properly thinking about types. Can do explicit type signatures which is great for teaching structured thinking. See More

Top Pro

Copyfree license (unlike Python and R)

Top Pro

Take advantages of matlab, R, Python

Specs

Site:julialang.org

Current stable version:0.6

Type Model:Dynamically Typed

See All

score 19 · Answer 3

R

Top Pro

Rich choice of high quality libraries

Be it data manipulation or plotting there is always a wide variety at CRAN available. See More

Top Con

Too slow

R is too slow to handle big data. See More

Top Pro

RStudio is a great IDE

It's free, lightweight, and easy to use, making working with R a pleasure. See More

Top Con

The library ecosystem (mostly CRAN but also github, BioConductor & others) is rather fragmented

There are over 10000 R packages and rising and there are many packages with overlapping functionality. It's not always clear which package is "best" for your needs. The situation is getting better thanks to Hadley Wickham who has rewritten a lot of core functionality from R base & popular but 'old school' packages in a few big alternative packages in a new more modern style and also more coherent (the so called 'Hadleyverse' or 'Tidyverse'). Together these form a framework analogous to what you have in Pydata for data science in Python or Spring for web applications in Java. See More

Top Pro

Open source

R is available under an open-source license, which means that anyone can download and modify the code. See More

Top Pro

ggplot2

Top Pro

knitr + Rmarkdown for (dynamic) documents

Top Pro

Lots of packages available

There are lots of different packages available that can be easily searched from the CRAN repo site and downloaded, or installed via the R command line interface. These packages are easy to include in a project or source file, and pertain to a wide variety of topics, from classification, social media analysis, and text processing to interactive 3d plotting and networks (including neural nets). See More

Top Pro

Very active and helpful community

The R community, especially on StackOverflow, is extremely helpful and reactive. It is really easy to find help when you are stuck with a certain problem, even as a complete beginner with basic problems. See More

Top Pro

Microsoft is now involved

Which means more high quality tooling (support for R in Visual Studio) & libs & integrations (e.g. R in SQLServer) & cloud solutions for scalability beyond single core & single host (R server , Spark via sparkR/sparklyr lib,... all on Azure supported by MS) . It also means a lot more 'mainstream' developers are going to be exposed to R (e.g. via SQLServer) and hence R is going to become a lot more popular outside of its traditional academic statistics niche See More

Top Pro

Shiny for web applications/dashboards

Top Pro

Has a wide range of options when it comes to IDEs/GUIs

Among the IDEs available there are several commercial applications as well as free and/or open-source ones, such as R Studio, which features syntax highlighting, project management capabilities, integrated terminal access, decent code completion and on-the-spot parameter hinting, graphical interfaces for package installation and such, and commendable extensibility/developer support. See More

Top Pro

There are so many R packages

There's always already a package for an algorithm described in a paper. See More

See All

score 19 · Answer 4

Rust

Top Pro

Catch errors at compile-time

Since Rust is statically typed, you can catch multiple errors during compile time. This is extremely helpful with debugging, especially compared with dynamically typed languages that may fail silently during runtime. See More

Top Con

Beginners could be stumped by some special concepts like Ownership, etc and spend extra time on it

Top Pro

Generics support

You don't have to write same array and dictionary classes hundreds and thousands times for strong type check by compiler. See More

Top Con

Hard to find learning resources or libraries

Because it's still a relatively new language, Rust does not enjoy a following as large as other languages/environments. Rust development has also been rather volatile for a long time during the beginning of the development of the language adding to this issue. Because of the small community, it's harder to find useful libraries for projects or any other kind of resource. See More

Top Pro

Built-in concurrency

Rust has built-in support for concurrency. See More

Top Con

Rust is a systems language not a data science language

Top Pro

Threads without data races

Unique ownership system guarantees a mutable datum to be owned and mutated by only one thread at a time, so there's no data race, and this guarantee is checked at compile time statically. Which means easy multi-threading. Of course, immutable data can be shared among multiple threads freely. See More

Top Pro

Compiles to machine code allowing for extra efficiency

Rust uses LLVM as a backend, among other things this allows Rust code to compile down to machine languages. This allows developers to write programs that run as efficiently as possible. See More

Top Pro

Support for macros

When you identify a part of your code which gets repeated often, which you cannot abstract using functions or classes, you can use Rust's built-in Macros. See More

Top Pro

Supports cross compilation for different architectures

Since Rust 1.8 you can install additional versions of the standard library for different targets using rustup/multirust. For example: $ rustup target add x86_64-unknown-linux-musl Which then allows for: $ cargo build --target x86_64-unknown-linux-musl See More

Top Pro

Makes developers write optimal code

Rust is a modern programming language written around systems. It was designed from the ground up this way. It's language design makes developers write optimal code almost all the time, meaning you don't have to fully know and understand the compiler's source code in order to optimize your program. Furthermore, Rust does not copy from memory unnecessarily, to give an example: all types move by default and not copy. Even references to types do not copy by default. In other words, setting a reference to another reference destroys the original one unless it's stated otherwise. See More

Top Pro

Easy to write understandable code

While not as verbose as Java, it still is much more verbose than languages like Go and Python. This means that the code is very explicit and easy to understand. See More

Top Pro

Developers of Rust consider data science support in Rust important and work on it

Specs

Site:rust-lang.org

Current stable version:1.26.0

License:MIT/Apache

See All

score 5 · Answer 5

Clojure

Top Pro

Incanter

Specs

Site:www.clojure.org

Current stable version:1.9

Price:Free, Open Source

GZipped size:3.7Mb, requires JAVA

score 7 · Answer 6

Scala

Top Con

Complex language

Not for beginners. See More

Top Pro

Algebraic types

Top Con

Can be intimidating for beginners

Scala is an industrial language. It brings functional programming to the JVM, but not with a "start small and grow the language" perspective, but rather a very powerful language for professional programmers. See More

Top Pro

Several data science libraries available (Scala or Java based)

Top Pro

Modern object-functional language that allows high level abstraction

Top Pro

Based on the JVM hence cross platform and much tooling

Top Pro

It's the language of Spark, the leading big data platform

Specs

Site:scala-lang.org

Current stable version:2.12.4

Price:Open Source (Free)

See All

score 2 · Answer 7

Haskell

Top Pro

Mathematical consistency

As Haskell lends itself exceedingly well to abstraction, and borrows heavily from the culture of pure mathematics, it means that a lot more code conforms to very high-level abstractions. You can expect code from vastly different libraries to follow the same rules, and to be incredibly self-consistent. It's not uncommon to find that a parser library works the same way as a string library, which works the same way as a window manager library. This often means that getting familiar and productive with new libraries is often much easier than in other languages. See More

Top Pro

Hand-writeable concise syntax

Conciseness of Haskell lets us to write the expression on the whiteboard or paper and discuss with others easily. This is a strong benefit to learn FP over other languages. See More

Top Pro

Easy to make DSL

You can handle data flow with your own DSL (Domain Specific Language). Haskell allow us easy to make DSL. See More

Top Pro

Easy to read

Haskell is a very terse language, particularly due to its type inference. This means there's nothing to distract from the intent of the code, making it very readable. This is in sharp contrast to languages like Java, where skimming code requires learning which details can be ignored. Haskell's terseness also lends itself to very clear inline examples in textbooks, and makes it a pleasure to read through code even on a cellphone screen. See More

Top Pro

Forces you to learn pure functional programming

It is pure and does not mix other programming paradigms into the language. This forces you to learn functional programming in its most pure form. This avoids you falling back on old habits and learn an entirely new way to program. See More

Specs

Site:https://www.haskell.org/

See All

score 1 · Answer 8

Lua

Top Pro

Great documentation

The official Lua documentation is very helpful and thorough. There are also a large number of online resources or books with lots of helpful information for beginners and advanced users alike. See More

Top Pro

Fast

Lua's performance compares very well to other languages, If performance needs to be further improved you can: Implement critical parts in C Use the LuaJIT compiler. The LuaJIT compiler is a drop in replacement for the stock compiler and provides significant performance improvements. From the overview page: LuaJIT speeds can rival code written in C. See More

score 2 · Answer 9

SAS

Add Video or Image

Answer 10

F#

Top Pro

Full access to all .NET libraries

Top Con

Only one worthy big ML lib available (Accord.net) and it's not even in F# but C#

Top Pro

Native support for Algebraic Data Types

Algebraic Data Types enables the creation of complex data structures in a very readable and reliable manner. See More

Top Con

No real data science/ML community in .NET & F#

Top Pro

High performance

Has the same performance profile as C#. See More

Top Pro

Easy access to R libraries via the R Type Provider

Top Pro

Simple ML-style expression based syntax

Specs

Site:http://www.fsharp.org

Current stable version:4.1

License:Apache 2.0

IDE Support:Visual Studio, Code, Rider

See All

score 2 · Answer 11

J

Top Pro

Free and open source

J is licensed under the GNU GPLv3. See More

Top Pro

High performance

Despite being interpreted, the array orientation gives the runtime good locality of reference for good cache performance and the repeated operators give it natural parallelism. See More

Top Pro

Array-oriented

J's verbs (operators) naturally work on multidimensional arrays with no need for manual looping. This makes J especially good at data analysis, and was the inspiration for Numpy's arrays. See More

Top Pro

Notation as a tool of thought

J's notation is very powerful. Once you learn it well, it allows you to to fit more into your limited human working memory. See More

See All

score 1 · Answer 12

Bash/awk/sed

Add Video or Image

Answer 13

Scheme

Add Video or Image

Answer 14

Common Lisp

Add Video or Image

score 1 · Answer 15

Java

Top Pro

Broad number of skilled developer avaialbe

Answer 16

Hy

Add Video or Image

score 1 · Answer 17

Scilab

Top Pro

Free

Top Con

Misses some basic functionality

For example like how to connect to a relational DB. See More

Top Con

Small community

Top Con

Doesn't have a lot of third party packages

See All

Answer 18

Nim

Top Pro

Great portability and optimization

Nim gains portability and optimization because it compiles to C, which offers a lot of choices from modern C compilers. See More

Top Pro

Cross-platform

Works on Windows, macOS, and Linux. See More

Specs

Site:https://nim-lang.org/

See All

Answer 19

APL

Add Video or Image

Answer 20

Kona

Add Video or Image

36 Options Considered

My Recommendation for Python

My Recommendation for Python

SciPy (Scientific Python library)

Slow

Rich set of libraries and extensions

Very limited support for functional programming which is important for data science

Clean syntax

Inelegant and messy language design

Full programming language, unlike R

Hard to debug other people's code

Numpy (Numerical Python library)

Does not teach you about data types

Portable

Language fragmentation

Pandas Library

Lack of statistical and machine learning algorithms

Lots of tutorials

Active and helpful community

Easy operator overloading

Databases

Has features of both high and low level language

Has many libraries for scientific computing, data mining and machine learning

Can be used in many domains

Very similar to pseudo-code

Cross-platform

Good documentation

SageMath

My Recommendation for Julia

My Recommendation for Julia

Multiple dispatch creates very clear code

Tooling & libs not yet up to par with others like R or Python

Nice regular syntax

Young language with limited support

Great standard REPL

Slow startup of programs due to compilation of the programs & (some of) its libs at startup

Fast computation

1-based array and column major

Easy Python interop

Plotting lib is not good

Lisp inspired macros

Though dynamically typed, can use type annotations to improve performance and make some errors easier to catch

Easy parallel computing

Strong dynamic typing

Copyfree license (unlike Python and R)

Take advantages of matlab, R, Python

My Recommendation for R

My Recommendation for R

Rich choice of high quality libraries

Too slow

RStudio is a great IDE

The library ecosystem (mostly CRAN but also github, BioConductor & others) is rather fragmented

Open source

ggplot2

knitr + Rmarkdown for (dynamic) documents

Lots of packages available

Very active and helpful community

Microsoft is now involved

Shiny for web applications/dashboards

Has a wide range of options when it comes to IDEs/GUIs

There are so many R packages

My Recommendation for Rust

My Recommendation for Rust

Catch errors at compile-time

Beginners could be stumped by some special concepts like Ownership, etc and spend extra time on it

Generics support

Hard to find learning resources or libraries

Built-in concurrency

Rust is a systems language not a data science language

Threads without data races

Compiles to machine code allowing for extra efficiency

Support for macros

Supports cross compilation for different architectures

Makes developers write optimal code

Easy to write understandable code

Developers of Rust consider data science support in Rust important and work on it

My Recommendation for Clojure

My Recommendation for Clojure

Incanter

My Recommendation for Scala