Code4Me provides automatic intelligent code completion based on large pre-trained language models. Code4Me predicts statement (line) completion and is available for both PyCharm (also other JetBrains IDEs) and Visual Studio Code. The code suggestions from Code4Me can be recognised by the logo in the suggestion menu. Code4Me automatically triggers on specific trigger characters or the user can prompt it manually by pressing the keybind. The keybind differs per IDE:
triggerSuggest command (likely CTRL + Space).Code4Me does not hinder native auto completion or other extensions. However, Code4Me could be hindered by other extensions. If this is the case, please disable the other (autocomplete) extensions. Please report the extension that causes trouble on our github repository by creating an issue.
The plugin is available on both marketplaces. You can download them by either clicking the link in your browser or looking up 'Code4Me' in the integrated marketplace of your IDE.
The Code4Me plugin can be found in the JetBrains Marketplace.
The Code4Me plugin can be found in the Visual Studio Code Marketplace.
The plugin works with a remote API in order to suggest autocompletions. After triggering the autocompletion (automatically or manually), the client's request is sent to the server. Running the ML-based model, the server returns a prediction in response to the client. The remote API requires a segment of the current document (close left context at the trigger point). This limited segment is sent to the backend server only for prediction and will not be stored on our server. For the purpose of evaluating the models' predictions, we only store the suggestion made by the ML-based models at each trigger point and compare it against the accepted prediction by users (collected after 30 seconds).
The plugin does not collect personal data nor the segment sent by default. The plugin does collect the following data:
One of our research goals is to identify where and why code autocompletion fails and how it can be improved. Hence, we would like to conduct a failure analysis on the completions provided by code4me. To this end, we need to study the code context fed to our models based on which the predictions are made. This context includes a number of code tokens before and after the completion point. This is an optional setting and this data will be collected anonymously. The data will only be stored for three months to perform the failure analysis. After this period, we completely remove this data.
Furthermore, Code4Me is in full compliance with the GDPR and all data is anonymous. The data collected will remain on the servers of TU Delft until the end of the study. By using Code4Me you give permission for the data collection.
Code4Me should automatically trigger on specific triggers points such as a period. Additionally, it can be manually triggered by the keybinds mentioned. If the manual trigger does not prompt a code suggestion it could be the case that the ML-based model did not find a completion. To make sure that this is not the case, please attempt to trigger the plugin with enough context (code).
If Code4Me still does not prompt autocomplete suggestions, please make sure that the suggestions have not dropped in ranking by scrolling down the suggestion list.
If there is no suggestion in the whole completion list than it is likely that another (autocomplete) extension is interferring with Code4Me. Please try disabling the other extensions.
If it still does not work, please create an issue on our github repository.
The source code for the IDE plugins and the API webserver can be found at our GitHub repository.