This dataset contains pairs <code changes, commit message> from top 1000-2000 Java projects in GitHub via GitHub Developer API. The structure of files is as follows: <repository-id>.<repository-name>/ # Each directory contains commits of one project. <commit-id>.<commit-sha>/ # Each directory contains one commit information, and # the commit-sha is the primary key of this commit in GitHub. # The children in this directory have 3 types: the file # "commit_msg", multiple directories "<filename>", and the file # "error". commit_msg # This file contains one line, representing the commit # message. <filename>/ # The name of this directory is the changed file name. patch # The code diffs of this <filename>, describing # which lines are added and which lines are # removed with some same lines context. <filename>/ patch error # This file contains the error message when crawling # from GitHub. If this file exists, the directories # <filename> will not exist.