Skip to content

README.md

Latest commit

 

History

History
32 lines (23 loc) · 3.18 KB

README.md

File metadata and controls

32 lines (23 loc) · 3.18 KB

Method Extraction

The first step is to extract the methods from the listed projects in setup.sh. Execute the following to extract the methods using 8 CPU cores:

bash scripts/extract_methods.sh method_extractor.log 8

This will create a log file which contains some stats about methods in logs/method_extractor.log. Moreover, it will store all methods and their metadata inside data/$project/unique_methods.jsonl. Each line in this file corresponds to a method and it comes in the following format:

{
    "index": "0", 
    "project": "commons-cli", 
    "file_path": "projects/commons-cli/src/main/java/org/apache/commons/cli/CommandLineParser.java", 
    "start_line": "65", 
    "end_line": "66", 
    "method": "    CommandLine parse(Options options, String[] arguments, boolean stopAtNonOption) throws ParseException;\n\n", 
    "tokens": [["CommandLine", ["source.java", "storage.type.java"]], ["parse", ["source.java", "meta.function-call.java", "entity.name.function.java"]], ["(", ["source.java", "meta.function-call.java", "punctuation.definition.parameters.begin.bracket.round.java"]], ["Options", ["source.java", "meta.function-call.java", "storage.type.java"]], ["options", ["source.java", "meta.function-call.java"]], [",", ["source.java", "meta.function-call.java", "punctuation.separator.delimiter.java"]], ["String", ["source.java", "meta.function-call.java", "storage.type.object.array.java"]], ["[", ["source.java", "meta.function-call.java", "punctuation.bracket.square.java"]], ["]", ["source.java", "meta.function-call.java", "punctuation.bracket.square.java"]], ["arguments", ["source.java", "meta.function-call.java"]], [",", ["source.java", "meta.function-call.java", "punctuation.separator.delimiter.java"]], ["boolean", ["source.java", "meta.function-call.java", "storage.type.primitive.java"]], ["stopAtNonOption", ["source.java", "meta.function-call.java"]], [")", ["source.java", "meta.function-call.java", "punctuation.definition.parameters.end.bracket.round.java"]], ["throws ParseException", ["source.java"]], [";", ["source.java", "punctuation.terminator.java"]]]
}

Attention Extraction

The second step is to extract attention weights from the methods in data/$project/unique_methods.jsonl. Execute the following to extract attention weights of codebert-base using GPU 0:

bash scripts/extract_attentions.sh attention_extractor.log codebert base 0

This will create a log file which contains some stats about methods in logs/attention_extractor.log. Moreover, it will store all methods and their attention weights inside data/$project/unique_methods_codebert-base_attnw.jsonl.

Attention Analysis

The third step is to analyze attention weights and determine Least Attended Tokens (LAT) and Least Attended Statements (LAS). Execute the following to analyze the attention weights of codebert-base and determine the LAT/LAS using 8 CPU cores with a threshold (k) of 10:

bash scripts/analyze_attentions.sh attention_analyzer.log codebert base 10 8

This will create a log file which contains some stats about methods in logs/attention_analyzer.log. Moreover, it will store all methods, attention weights, and LAS/LAT inside data/$project/unique_methods_codebert-base_las_lat.jsonl.