Depp TabNine

2019: Output of the tool

In July, 2019 it became known of release of the tool using artificial intelligence for fast writing of the Deep TabNine code. It supports 22 programming languages, including Python, JavaScript, Java, C ++, C, PHP, Go, C#, Ruby, Objective-C, Rust and Apple Swift. The program is written on Rust.

Judging by information from the project website, it is developed by TabNine Inc company. Behind it there is a student of faculty of information science of the University of Waterloo Jacob Jackson.

The AI tool which accelerates writing of the code in 22 programming languages is provided

Deep TabNine represents a plug-in for code editors which analyzes the lines made by the programmer and  then generates options for end of each "phrase". Then the person selects  the most suitable methods from the list to continue the code.

For increase in efficiency of automatic filling of fragments of the code the program was trained at a basis of 2 million files placed on the GitHub portal. The algorithm is based on neural network GPT-2 in which development the non-profit research company OpenAI is engaged. GPT-2 is intended for natural languag processing.

The neuronet was adapted to understanding of the code and learned to generate the connected texts  and also to supplement them  with a missing word and combinations.

According to Dzheksn, the artificial intelligence studies the used syntax in time of writing of the code and foretells when the programmer calls a method of an object or the list. It can even generate names of objects, using documentation written in a natural language.

By July 25, 2019 Deep TabNine is at a beta testing stage in a cloud. Anyone can test the tool, having sent the request.

Jackson says that the command works on the cut-down model of machine learning which can work on the computer of certain developers. Besides, creation of corporate versions in which the model will be trained according to requirements of the specific company is expected.^[1]

Notes