How wuyifan18/adaparser works
Project Type and Purpose
• This is an open-source research tool for log parsing experiments.
• It provides two main functions:
– Parsing logs with an LLM-based parser (AdaParser)
– Evaluating the parsed results (accuracy metrics)
Main Components
A. Data Layer
– full_dataset/ (unzipped Loghub-2.0 raw logs)
– sampled_examples_* subdirectories (pre-sampled log snippets)
B. Parsing Service
– benchmark/logparser/AdaParser/AdaParser.py (core parsing logic)
– Trie.py, post_process.py, utils.py (supporting code)
– openai_key.txt → holds API key for invoking OpenAI LLM
C. Orchestration (CLI Entry Point)
– benchmark/evaluation/AdaParser_eval.py
• Parses CLI args: --model, --log_ratio
• Loads sampled logs, invokes AdaParser modules
• Calls evaluation utilities
D. Evaluation Service
– benchmark/evaluation/utils/
• PA_calculator.py (Parsing Accuracy)
• GA_calculator.py (Grouping Accuracy)
• template_level_analysis.py, postprocess.py
• evaluator_main.py (ties calculators together)
E. Persistence / Output
– result/ directory (not in repo, created at runtime)
– Stores parsed logs, metrics, and summary reports
F. External Dependency
– OpenAI API (LLM endpoint)
Relationships & Data Flows
1. CLI invocation
→ reads openai_key.txt
→ reads sampled logs from full_dataset
→ calls AdaParser (Parsing Service)
2. AdaParser
→ constructs prompts (self-generated in-context examples)
→ sends to OpenAI API → receives parsed templates
→ applies post_process.py & Trie matching
3. Parsed output
→ fed into Evaluation Service (GA & PA calculators)
4. Evaluation results & artifacts
→ written to result/
Architectural Patterns & Principles
• Layered architecture: Data → Parsing → Evaluation → Persistence
• Single‐process, CLI‐driven orchestration
• Clear separation of concerns:
– “logparser” package for parsing logic
– “evaluation” package for metrics
Technologies & Frameworks
• Python 3.8+
• OpenAI Python SDK (via requests or openai library)
• Standard Python packages (json, argparse, etc.)
• Requirements.txt lists specific deps
Detail Checklist
[ ] Callouts for key scripts (AdaParser_eval.py, AdaParser.py, GA_calculator.py)