Method for carrying out JavaScript type inference based on deep learning

Abstract

The invention relates to deep learning in the field of artificial intelligence, in particular to source code learning. A method comprises the steps of data collection and processing, model construction, model training, model evaluation and type inference. The step of data collection and processing comprises: firstly, downloading a certain amount of source codes on Github; screening source codes with rich types as a final data set; then converting data into a format in which tokens and types are aligned and matched, meanwhile, generating token and type vocabulary libraries, and finally, expressing the source code as a data format, such as a vector, suitable for learning by utilizing token-type mapping. The construction of the model comprises the steps of firstly determining the type of theneural network, then determining the number of layers of the neural network, and finally determining the number of neurons of each layer. The training of the model comprises the steps of tracking a loss function value and a classification error, and updating model parameters until the model with relatively high accuracy is obtained. The model evaluation comprises statistics of the accuracy and consistency of the model. The type inference comprises the steps of loading the trained model with higher accuracy, marking the type deduction result behind the corresponding identifier, and finally outputting the type inference result in the form of a file. The process is shown in Figure 1.

Description

technical field [0001] The invention relates to deep learning in the field of artificial intelligence, in particular to a learning of source codes. Specifically, after processing the source code, it is used as the input of the neural network, and then trained to obtain a model with high accuracy, so as to perform JavaScript type inference. Background technique [0002] In recent years, deep learning has been widely used in speech recognition, machine translation, automatic driving and other fields. At present, the learning of source code has also attracted people's attention. The advantage of deep learning is that it can extract highly complex features through deep neural networks, while machine learning needs to define features by itself. At present, deep neural networks have achieved success in a series of natural language tasks. Unlike natural language, programs contain rich, clear, and complex structural information. Traditional natural language processing models (NLP,...

AI Technical Summary

Problems solved by technology

There are also some other code inspection tools, but these tools have some shortcomings, such as the inability to detect type errors such as reference types and custom types, so type inference for JavaScript is necessary

Images