Top 5 programming languages ​​recommended for AI development! Introduced in ranking format

1. Ranking of recommended programming languages ​​suitable for AI (artificial intelligence) development

Demand for AI has been increasing in recent years, but at the same time as the opportunity to use it, the opportunity to actually develop it is becoming familiar.

However, many people may not know which programming language to develop in. This time, we will introduce recommended languages ​​suitable for AI development in a ranking format.

1.1 Selection criteria

First, I will explain the points to consider when selecting a programming language in AI development.

1.1.1 Grammar clarity

In the first place, in order to acquire the skills to develop AI at a practical level in order to utilize AI, knowledge of mathematics and data analysis is required as well as programming languages ​​for artificial intelligence development .

Among the mathematics required for machine learning including deep learning, the basic ones are “derivative”, “linear algebra”, and “probability statistics” .

Machine learning is performed by minimizing and maximizing a function called an objective function to determine parameters that behave as desired. And, in order to solve the problem of minimization of the objective function, knowledge of differentiation and linear algebra is required.

Learning programming and math in parallel can be time consuming, costly and unreasonable. If you are new to programming and are not confident in your mathematical knowledge of linear algebra and statistics, you should choose a programming language that has a simple grammar and is suitable for beginners.

Once you have some skill and knowledge, it is recommended that you learn a language that is more extensible and runs faster.

1.1.2 Activeness of the development community

Compared to major languages, programming languages ​​that are minor or difficult have a smaller number of participants in the domestic development community, and the pace of study sessions is sparse. In addition, it tends to be difficult to find blog articles and posts that summarize error resolution methods that are helpful during development.

For smooth development, it’s a good idea to choose a major programming language as your main programming language .

1.1.3 Many frameworks supported

For example, TensorFlow supports so many languages ​​that you can choose from Python and C, Java, Go, C ++, and more. You can also use “TensorFlow.js” with Nodejs and JavaScript.
On the other hand, Pytorch supports Python, C ++, and CUDA.

If you want to develop with both TensorFlow and Pytorch, you will inevitably have two choices, Python or C ++.

One way to do this is to first think about which framework you want to use, then back-calculate which language the framework supports to determine the language .

1.1.4 Large number of projects

If you are also considering accepting orders for AI development work as a freelancer, the number of projects that can be handled in that language is also important.
In particular, if you become independent without a large track record in the field of artificial intelligence, it is necessary to receive orders not only for artificial intelligence but also for Web application development and tool development in parallel to create financial margin and order record. There may be.

Therefore, “Is it a language with many projects related to artificial intelligence development?” “Even if there are not so many projects with artificial intelligence, is it likely that if you have that language, you will be able to receive orders for projects such as application development?” It is recommended that you do some preliminary research before learning the language.

If the language is too minor, it may not be possible to win a deal, so you need to choose carefully.

1.2 [5th place] MATLAB

A programming language used in numerical analysis software developed by MathWorks.

It is said to be suitable for data analysis, algorithm development, and model creation, and can reduce the time required for data preprocessing and enable quick visualization of data. In addition, you can easily apply general analysis algorithms, etc., or you can write your own code for analysis.

It can also be deployed in C language and C ++ code, and can be applied to multiple platforms and commercialized.

MATLAB, on the other hand, is not an open source language, so you have to buy a package. In addition, because it is a very specialized language, it is not often used by companies outside of niche fields such as data analysis, engineering, and manufacturer development, so it can be said that there are high hurdles for beginners to use.

1.3 [4th place] C ++

An object-oriented programming language developed by Nokia Bell Labs computer scientist Bjarne Stroustrup in 1983.

A major feature of C ++ is its high processing speed. Since an executable file that is extremely close to machine language is generated and executed, processing is performed faster than in other languages. You can also access C code from C ++ and have direct access to hardware resources and system functions. In processing that handles huge and heavy data such as big data and AI learning, the processing speed may have a large effect on business productivity, so it is a point that must be taken into consideration especially in large-scale projects, and the introduction of C ++ The merit is great.

By using C ++, you can create your own Python library, and you can do things that existing libraries cannot do.

Once learned, it is a language with great merits, but the bottleneck is that you have to learn classes, object orientation, memory management, metaprogramming with templates, etc., and the learning cost is high.

1.4 [3rd place] Julia

It is attracting attention as a free language for scientific and technological calculations, and is characterized by the ability to easily create programs for numerical calculations such as matrix calculations. It is a relatively new language, and is attracting attention as a language that has speed like C language, description familiar with mathematical formulas like MATLAB, and versatility like Python.

It is also possible to call Python functions and C language code from within Julia, and one of the merits is that it is compatible with other languages. Execution speed is also faster than Python and R language.

It is a language that is attracting attention as a language for machine learning, numerical analysis, and big data, but as of October 2020, it is lagging behind Python and R languages ​​due to the abundance of libraries and the number of projects. In terms of practicality and development cost, the aspect of “developing language” is still strong, so it is a language that I would like to expect in the future.

1.5 [2nd place] R

Developed in 1995 by scientist Ross Ihaka and Robert Gentleman of the University of Auckland, it is a programming language suitable for data analysis and statistics. Originally a language for academics and research, it has been increasingly used for business purposes in recent years. Prior to the advent of the R language, high-priced software for professionals was the major source of statistical processing, and the free software R language was invaluable.

The feature is the richness of “packages” that correspond to Python’s “libraries”. This enables flexible data analysis and visualization such as financial engineering, time series analysis, machine learning, data mining, and bioinformatics, regardless of academic field or industry. It is also possible to immediately introduce and execute algorithms developed by developers in various regions.

The disadvantage is that it was originally intended for academics and research, so it has a strong aspect of specializing in data analysis and mathematics, and does not have the flexibility and versatility of Python. Since there is a framework for web application development called Shiny, it is possible to develop applications, but since there are few Japanese documents, it has not left its minor position.

1.6 [No. 1] Python

A scripting language developed by Guido van Rossum and first released in 1991. Emphasis is placed on the “readability” and “easiness of understanding” of the program, and it is designed with a really simple grammatical structure, “only one way to express one thing”.

Python is characterized by its ease of use over execution performance. In the first place, the language specifications themselves are not suitable for scientific and technological calculations. However, the flexible architecture also allows high-speed processing developed in Fortran and C to be embedded in Python and extended for use. By leveraging these extensions, even non-programming researchers can use easy-to-learn Python to perform scientific and technological calculations.

The disadvantage is that it is a scripting language and does not run fast. For large-scale data analysis projects where execution speed is severely questioned, it can be said that C ++ or MATLAB, which sells high-speed processing speed, is more suitable.

2. [By application] Recommended AI development language

The situation and environment for AI development will vary from project to project.
In this chapter, we will introduce recommended AI development languages ​​for each situation.

2.1 [For beginners] Python / R

In general, scripting languages ​​are said to be for beginners because their grammar is relatively low and easy to execute.
Both Python and R are scripting languages ​​and open source, so you can easily try them from installation to execution for free. In addition, since there are many libraries (packages) and many Japanese documents, even people with little expertise in programming, artificial intelligence, and data analysis can easily start analysis .

2.2 [Language to learn next to Python / R] Julia

Julia, whose grammar is close to that of a general-purpose language and whose execution speed is fast, can call functions in C, Fortran, and Python.

If you are learning Julia , you can use it with a language such as Python to build a program that can process faster. It can be said that it is a language that has the potential to develop into one of the options together with C ++ in large-scale analysis where execution speed is severely questioned.

Since it is a new language and the library is still lacking, it is recommended to study Julia with a certain coding skill after laying the foundation of AI development in Python and R language .

2.3 [I want to extend the Python library] C ++

It is possible to call the process written in C ++ from Python by using the extension module in Python. By utilizing this function, you can expect the effect of increasing the processing speed of the program.

By utilizing C ++, you will be able to improve the usability of the Python library.

2.4 [I want to put the highest priority on execution speed] C ++

C ++ is a language with direct access to hardware resources and system functions. It can be said that it is one of the languages ​​that can perform fairly high-speed processing because it generates and executes an executable file that is extremely close to machine language.

It is a language that should be actively considered for introduction in projects where processing speed is an issue .

There is also room for consideration in how to use it to improve the processing speed by calling C ++ processing from Python.

3. Summary

Every language has its advantages and disadvantages, such as ease of use, execution speed, and future potential. Which points should be prioritized depends on the situation of the developer and the project.
It can be said that it is important in AI development to research various programming languages ​​and select them in light of development requirements and one’s own skills.