Amazon Technologies, Inc.

Srinivasan Sengamedu Hanumantha Rao

Neela Sawant

Hoan Nguyen

Omer Tripp

Pranav Garg

Russell Reas

Shishir Sethiya

Anton Emelyanov

Jasmeet Chhabra

Qiang Zhou

Hakimuddin Hanif

Prakash Mandayam Comar

Sahil Sareen

Yinglong Wang

Bibaswan Kumar Chatterjee

Alok Dhamanaskar

Himani Khanduja

Harshit Gupta

Xiaoxin Zhao

Techniques for management of sensitive data using static code analysis are described. A method of management of sensitive data using static code analysis includes obtaining a representation at least a portion of code, statically analyzing at least the portion of code to generate one or more candidate vectors based at least on one or more patterns, sending the one or more candidate vectors to a sensitive data model, and receiving an inference response indicating, for each of the one or more candidate vectors, whether at least a portion of the candidate vector includes sensitive data and a corresponding confidence score.

Management of sensitive data using static code analysis

Techniques for performing machine learning-based program analysis using synthetically generated labeled data are described. A method of performing machine learning-based program analysis using synthetically generated labeled data may include receiving a request to perform program analysis on code, determining a first portion of the code associated with a first error type, sending the first portion of the code to an endpoint of a machine learning service associated with an error detection model to detect the first error type, the error detection model trained using synthetically generated labeled data, and receiving inference results from the error detection model identifying one or more errors of the first error type in the first portion of the code.

Machine learning-based program analysis using synthetically generated labeled data

Techniques for a code reviewer service to provide recommendations on source code are described. A code reviewer service may run rules and/or machine learning models to provide the recommendations. A machine learning model may identify one or more predicted issues of source code, and the code reviewer service may provide one or more recommendations based at least in part on the one or more predicted issues. Code reviewer service may allow a pull request for a code repository to trigger the generation of recommendations for the source code in the repository. The recommendations may be posted on the pull request.

Customizable service for recommendations on source code quality

Techniques for generating rules from documentation are described. For example, a method for generating rules may include generating one or more rules from documentation by: extracting a plurality of chunks from the documentation, inferring one or more candidate rules from the extracted chunks, mining the inferred one or more candidate rules to determine at least one of the one or more candidate rules is to be included in rule generation, classifying the at least one mined one or more candidate rules as one or more rules, and extracting information to generate the one or more rules.

Mining software specification from online documentation

Techniques for generating rules from documentation are described. For example, a method for generating rules may include generating one or more templates containing patterns and anti-patterns from code of one or more documents, wherein a pattern captures a good coding practice as detailed in the documents and an anti-pattern is a proper subset of a pattern that does not include a construct described in the documents as being a recommended and/or required usage; constructing one or more graphs from the one or more templates; mining the constructed one or more graphs to find target sub-graphs which do not contain bugs, wherein a bug occurs when an anti-pattern matches, but a pattern does not match; comparing the target sub-graphs to a representative dataset to remove target sub-graphs that violate good usage; and codifying the sub-graphs that represent good usage.

Codifying rules from online documentation

Techniques for unsupervised learning of embeddings on source code from non-local contexts are described. Code can be processed to generate an abstract syntax tree (AST) which represents syntactic paths between tokens in the code. Once the AST(s) have been generated, the paths in the AST(s) can be crawled to identify terminals (e.g., leaf nodes in the AST) and paths between terminals can be identified. The pairs of tokens identified at the ends of each path can then be used to generate a cooccurrence matrix. For example, if X number of unique terminals are identified, a matrix of size X by X can be generated to indicate a frequency at which pairs of terminals cooccur. This cooccurrence matrix can then be used as input to existing techniques for learning vector-space embeddings, such as word2vec, GloVe, Swivel, etc.

Techniques for unsupervised learning embeddings on source code tokens from non-local contexts

A transformed data set corresponding to a machine learning classifier's training data set is generated. Each transformed record contains a modified version of a corresponding training record, as well as the prediction made for the training record by the classifier. A set of explanatory rules is minded from the transformed data set, with each rule indicating a relationship between the prediction and one or more features corresponding to the training records. From among the rule set, a particular matching rule is selected to provide an easy-to-understand explanation for a prediction made by the classifier for an observation record which is not part of the training set.

Explainers for machine learning classifiers

Techniques for suggesting a name from one or more code files are described. An exemplary method includes receiving a request to suggest one or more names for a name in a code file; determining one or more names based on existing names in one or more code files using one or more abstract syntax trees (ASTs) for the one or more code files; and outputting the determined one or more names as a name suggestion that comprises novel sequences of sub-tokens of existing names of the one or more code files.

Using an abstract syntax tree for generating names in source code

The effect of intent bias on content performance can be determined in order to provide more relevant content in response to a query or other opportunity. Performance data can include the frequency with which an action, such as a purchase, occurs in response to an instance of the content being displayed. An intent bias model can be trained using the performance data for two or more intents, such as an action intent and an explore intent. Once the intent bias for an offer is determined, a normalized performance value can be obtained that does not include the effects of the bias. The normalized values can be used to select and place content based on actual performance.

Growing community of inventors

Seattle, WA, United States of America

Srinivasan Sengamedu Hanumantha Rao