Computer code refactoring
Systems and methods are provided for automated computer code editing. The method includes training a code-editing neural network model using a corpus of code editing data samples, including the pre-editing samples and post-editing samples, and parsing the pre-editing samples and post-editing samples into an Abstract Syntax Tree (AST). The method further includes using a grammar specification to transform the AST tree into a unified Abstract Syntax Description Language (ASDL) graph for different programming languages, and using a gated graph neural network (GGNN) to compute a vector representation for each node in the unified Abstract Syntax Description Language (ASDL) graph. The method further includes selecting and aggregating support samples based on a query code with a multi-extent ensemble method, and altering the query code iteratively using the pattern learned from the pre- and post-editing samples.
1. A computer implemented method for automated computer code editing, comprising:
training a code-editing neural network model using a corpus of code editing data samples, including the pre-editing samples and post-editing samples;
parsing the pre-editing samples and post-editing samples into an Abstract Syntax Tree (AST);
using a grammar specification to transform the AST tree into a unified Abstract Syntax Description Language (ASDL) graph for different programming languages;
using a gated graph neural network (GGNN) to compute a vector representation for each node in the unified Abstract Syntax Description Language (ASDL) graph;
selecting and aggregating support samples based on a query code with a multi-extent ensemble method; and
altering the query code iteratively using the code-editing neural network model and the pattern learned from the pre- and post-editing samples.
2. The computer implemented method of claim 1 , wherein the pre-editing samples and post-editing samples are combined using the multi-extent ensemble method.
3. The computer implemented method of claim 2 , wherein a query code snippet corpus data includes a set of query code snippets for each edit type.
4. The computer implemented method of claim 3 , further comprising inputting the query code as an initial input for code editing.
5. The computer implemented method of claim 4 , wherein the query code is a set of support code snippets provided as edit samples for both the pre-edit and post-edit code snippets in training.
6. The computer implemented method of claim 5 , further comprising extending the original AST tree by adding bidirectional edges between parent and child nodes.
7. The computer implemented method of claim 6 , further comprising utilizing mean-pool to represent a whole graph.
8. The computer implemented method of claim 7 , further comprising computing an activation for each of the query and support samples.
9. The computer implemented method of claim 8 , wherein the activations of the query and support samples are given by:
σ n q :=max{φ( f q ( z n q ), f s ( z i s ))} i=1 N s and σ n s :=max{φ( f s ( z n s ), f q ( z i q ))} i=1 N q ,
respectively.
10. The computer implemented method of claim 9 , wherein the graph representation of a query code graph and a support code graph is generated by:
t
λ
q
:=
∑
n
=
1
N
q
∑
k
=
1
K
1
K
σ
n
,
λ
q
k
z
n
q
and
t
λ
S
:=
∑
n
=
1
N
s
σ
n
,
λ
s
z
n
s
,
respectively.
11. A computer system for automated computer code editing, comprising:
one or more processors;
a computer memory operatively coupled to at least one of the one or more processors;
a computer display operatively coupled to the computer memory and the at least one of the one or more processors, wherein the computer memory contains instruction code that can cause the computer system to perform the operations of:
training a code-editing neural network model using a corpus of code editing data samples, including the pre-editing samples and post-editing samples;
parsing the pre-editing samples and post-editing samples into an Abstract Syntax Tree (AST);
using a grammar specification to transform the AST tree into a unified Abstract Syntax Description Language (ASDL) graph for different programming languages;
using a gated graph neural network (GGNN) to compute a vector representation for each node in the unified Abstract Syntax Description Language (ASDL) graph;
selecting and aggregating support samples based on a query code with a multi-extent ensemble method; and
altering the query code iteratively using the code-editing neural network model and the pattern learned from the pre- and post-editing samples.
12. The computer system of claim 11 , wherein the pre-editing samples and post-editing samples are combined using the multi-extent ensemble method.
13. The computer system of claim 12 , wherein a query code snippet corpus data includes a set of query code snippets for each edit type.
14. The computer system of claim 13 , further comprising instruction code that can cause the computer system to perform the operations of inputting the query code as an initial input for code editing.
15. The computer system of claim 14 , wherein the query code is a set of support code snippets provided as edit samples for both the pre-edit and post-edit code snippets in training.
16. The computer system of claim 15 , further comprising instruction code that can cause the computer system to perform the operations of extending the original AST tree by adding bidirectional edges between parent and child nodes.
17. The computer system of claim 16 , further comprising instruction code that can cause the computer system to perform the operations of utilizing mean-pool to represent a whole graph.
18. The computer system of claim 17 , further comprising instruction code that can cause the computer system to perform the operations of computing an activation for each of the query and support samples.
19. The computer system of claim 18 , wherein the activations of the query and support samples are given by:
σ n q :=max{φ( f q ( z n q ), f s ( z i s ))} i=1 N s and σ n s :=max{(φ( f s ( z n s ), f q ( z i q ))} i=1 N q ,
respectively.
20. The computer system of claim 19 , wherein the graph representation of a query code graph and a support code graph is generated by:
t
λ
q
:=
∑
n
=
1
N
q
∑
k
=
1
K
1
K
σ
n
,
λ
q
k
and
z
n
q
t
λ
S
:=
∑
n
=
1
N
s
σ
n
,
λ
s
z
n
s
,
respectively.