Code adaptation through deep learning

A method comprising:

• obtaining source code samples, wherein each of the source code samples includes source code in which a defined variable name is replaced with a masked value to generate a training instance for learning an association between the masked value and the defined variable name, such that multiple defined variable names and multiple masked values are accessible;
• extracting an extended context for each of the source code samples, the extended context including one or more code elements outside of a local scope of a source code snippet within said each source code sample, such that multiple extended contexts are accessible;
• generating a training dataset based on the extended contexts and the masked values for the defined variable names;
• and
• providing the training dataset to facilitate training a deep learning model, resulting in the deep learning model learning an association between a given masked value and a given defined variable name,
• wherein the trained deep learning model is configured, in response to target source code that includes an external source code snippet having an undefined variable name, to generate a particular defined variable name corresponding to the undefined variable name and to modify the target source code by replacing the undefined variable name with the particular defined variable name.

One or more hardware storage devices that store instructions that are executable by one or more processors to cause the one or more processors to:

• obtain source code samples, wherein each of the source code samples includes source code in which a defined variable name is replaced with a masked value to generate a training instance for learning an association between the masked value and the defined variable name, such that multiple defined variable names and multiple masked values are accessible;
• extract an extended context for each of the source code samples, the extended context including one or more code elements outside of a local scope of a source code snippet within said each source code sample, such that multiple extended contexts are accessible;
• generate a training dataset based on the extended contexts and the masked values for the defined variable names;
• and
• provide the training dataset to facilitate training a deep learning model, resulting in the deep learning model learning an association between a given masked value and a given defined variable name,
• wherein the trained deep learning model is configured, in response to target source code that includes an external source code snippet having an undefined variable name, to generate a particular defined variable name corresponding to the undefined variable name and to modify the target source code by replacing the undefined variable name with the particular defined variable name.