Skip to content

Latest commit

 

History

History
98 lines (71 loc) · 3.58 KB

storages.md

File metadata and controls

98 lines (71 loc) · 3.58 KB
Raw

Storages

The storage defines how the ASTs should be saved on disk. For now, astminer supports several tree-based and path-based storage formats.

astminer also knows how to find the structure of the dataset and can save trees or path contexts in the appropriate holdout folders. (train, val and test). If the data is not split, all trees will be saved in the data folder. Description files for trees or paths will be saved along with holdouts in the same outputPath directory.

Storage config classes are defined in StorageConfigs.kt.

Tree formats

CSV

Saves the trees with labels to a comma-separated file. Each tree is encoded to a single line using parentheses sequences.

name: csv AST

Dot

Saves each tree in a separate file using the dot syntax. Along with dot files, this storage also saves description.csv with a mapping between files with trees, source files, and labels.

name: dot AST

JSON lines

Saves each tree with its label in the JSON lines format inspired by the 150k Python dataset.

name: json AST

In this format, each line represents an AST with its label, path, and all vertices:

{
  "label": "1.java",
  "path": "src/test/resources/examples/1.java",
  "ast": [
    { "token": "EMPTY", "typeLabel": "CompilationUnit", "children": [1] },
    { "token": "class", "typeLabel": "TypeDeclaration", "children": [2, 3, 4] },
    ...
  ]

Path-based representations

Path-based representation was introduced by Alon et al.. It is used in popular code representation models such as code2vec and code2seq.

Code2vec

Extracts paths from each AST. The output is stored in 4 files:

  1. node_types.csv contains numeric IDs and corresponding node types with directions (up/down, as described in this paper by Uri Alon et al.).

  2. tokens.csv contains numeric IDs and corresponding tokens.

  3. paths.csv contains numeric IDs and AST paths in the form of space-separated sequences of node type IDs.

  4. path_contexts.c2s contains the labels and sequences of path-contexts (each representing two tokens and a path between them). This file is generated for every holdout.

    Each line in path_contexts.c2s starts with a label followed by a sequence of space-separated triples. Each triple contains comma-separated IDs of the start token, path, and end token.

name: code2vec
maxPathLength: 10
maxPathWidth: 2
maxTokens: 1000 # can be omitted
maxPaths: 1000 # can be omitted
maxPathContextsPerEntity: 200 # can be omitted

Code2seq

Extracts paths from each AST and save in the code2seq format. The output is path_context.c2s file, which is generated for every holdout. Each line starts with a label followed by a sequence of space-separated triples. Each triple contains comma-separated IDs of the start token, path node types, and end token.

To reduce memory usage, you can enable the nodesToNumbers option. If nodesToNumbers is set to true, all types are converted into numbers and node_types.csv with the node-number vocabulary is added to the output files.

name: code2seq
length: 10
width: 2
maxPathContextsPerEntity: 200 # can be omitted
nodesToNumbers: true # can be omitted

length stands for the maximum length of a path inclusively; width stands for the maximum distance between the children of the least common ancestor of a path.