Distributed vectorized representations of source code commits

1. A method comprising:

receiving a source code commit comprising source code, time information, and a label associated based upon a machine learning target task regarding a security vulnerability in the source code, a fix to a bug in the source code, or a new feature in the source code;

referencing the time information to split the source code into pre-change source code and post-change source code;

converting the pre-change source code into a first code representation;

converting the post-change source code into a second code representation;

generating a plurality of first particles from the first code representation, a particle being a smallest element in a given form of code representation;

generating a plurality of second particles from the second code representation;

comparing the plurality of first particles and the plurality of second particles to create a delta;

transforming the delta into a first commit vector by referencing an embedding matrix to numerically encode particles of the delta;

classifying the first commit vector according to the machine learning target task; and

storing the first classified commit vector in a non-transitory computer readable storage medium.

2. A method as in claim 1 wherein the first code representation and the second code representation use text tokens.

3. A method as in claim 1 wherein the first code representation and the second code representation use a graph model.

4. A method as in claim 3 wherein the graph model is an Abstract Syntax Tree (AST).

5. A method as in claim 4 wherein the graph model is a combination of graph representations.

6. A method as in claim 1 wherein the source code commit is part of a training corpus having a first volume, the method further comprising:

generating a second commit vector from a pre-training corpus having second volume larger than the first volume;

classifying the second commit vector according to a pre-training task related to the machine learning target task; and

storing the classified second commit vector in the non-transitory computer readable storage medium.

7. A method as in claim 6 further comprising using backpropagation to train a classifier to perform the classification of the first commit vector.

8. A method as in claim 1 further comprising:

prior to the receiving, associating the label with the source code commit according to the machine learning target task.

9. A method as in claim 1 wherein:

the non-transitory computer readable storage medium comprises an in-memory database; and

the transforming is performed by an in-memory database engine of the in-memory database.

10. A non-transitory computer readable storage medium embodying a computer program for performing a method, said method comprising:

receiving a source code commit comprising source code, time information, and a label associated based upon a machine learning target task regarding a security vulnerability in the source code, a fix to a bug in the source code, or a new feature in the source code;

referencing the time information to split the source code into pre-change source code and post-change source code;

converting the pre-change source code into a first graph representation;

converting the post-change source code into a second graph representation;

generating a plurality of first particles from the first graph representation, a particle being a smallest element in a given form of code representation;

generating a plurality of second particles from the second graph representation;

comparing the plurality of first particles and the plurality of second particles to create a delta;

transforming the delta into a first commit vector by referencing an embedding matrix to numerically encode particles of the delta;

classifying the first commit vector according to the machine learning target task; and

storing the first classified commit vector in a non-transitory computer readable storage medium.

11. A non-transitory computer readable storage medium as in claim 10 wherein the first graph representation is:

an Abstract Syntax Tree (AST);

a Control-Flow Graph (CFG);

a Data-Flow Graph (DFGs); or

a combination of at least two of an AST, a CFG, and a DFG.

12. A non-transitory computer readable storage medium as in claim 10 wherein the source code commit is part of a training corpus having a first volume, and the method further comprises:

generating a second commit vector from a pre-training corpus having second volume larger than the first volume;

classifying the second commit vector according to a pre-training task related to the machine learning target task; and

storing the classified second commit vector in the non-transitory computer readable storage medium.

13. A non-transitory computer readable storage medium as in claim 12 wherein the method further comprises:

using backpropagation to train a classifier to perform the classification of the first commit vector.

14. A non-transitory computer readable storage medium as in claim 10 wherein the method further comprises:

prior to the receiving, associating the label with the source code commit according to the machine learning target task.

15. A computer system comprising:

one or more processors;

a software program, executable on said computer system, the software program configured to cause an in-memory database engine of an in-memory database to:

receive a source code commit comprising source code, time information, and a label associated based upon a machine learning target task regarding a security vulnerability in the source code, a fix to a bug in the source code, or a new feature in the source code;

reference the time information to split the source code into pre-change source code and post-change source code;

convert the pre-change source code into a first code representation;

convert the post-change source code into a second code representation;

generate a plurality of first particles from the first code representation, a particle being a smallest element in a given form of code representation;

generate a plurality of second particles from the second code representation;

compare the plurality of first particles and the plurality of second particles to create a delta;

transform the delta into a first commit vector by referencing an embedding matrix to numerically encode particles of the delta;

classify the first commit vector according to the machine learning target task; and

store the first classified commit vector in the in-memory database.

16. A computer system as in claim 15 wherein first code representation and the second code representation use text tokens.

17. A computer system as in claim 15 wherein the first code representation and the second code representation use a graph model.

18. A computer system as in claim 17 wherein the graph model is:

an Abstract Syntax Tree (AST);

a Control-Flow Graph (CFG);

a Data-Flow Graph (DFGs); or

a combination of at least two of an AST, a CFG, and a DFG.

19. A computer system as in claim 15 wherein the source code commit is part of a training corpus having a first volume, and the in-memory database engine is further configured to:

generate a second commit vector from a pre-training corpus having second volume larger than the first volume;

classify the second commit vector according to a pre-training task related to the machine learning target task; and

store the classified second commit vector in the non-transitory computer readable storage medium.

20. A computer system as in claim 19 wherein the in-memory storage engine is further configured to use backpropagation to train a classifier to perform the classification of the first commit vector.