Causal Discovery on DWDClimate
88F1 ScoreMATMCD
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MATMCDLLM=GPT-4o mini, Top-K Guess Reasoning=K=12024.12 | 88 | 75 | 3 | 4 | 11 | |
| Auto Opt. (Ours)Time (s)=169, Cost ($)=2.32, Description=Stage 2 result after automatic skill iteration2026.04 | 70.1 | — | — | — | — | |
| MATMCD (MAS)Time (s)=1852, Cost ($)=2.792026.04 | 66.3 | — | — | — | — | |
| MAS CompilerTime (s)=149, Cost ($)=1.462026.04 | 64.2 | — | — | — | — | |
| LLM-KBCI-RALLM=GPT-4o mini, Framework=ReAct2024.12 | 60 | 75 | 3 | 4 | 11 | |
| Adaptive Skill (Ours)Time (s)=137, Cost ($)=1.772026.04 | 59.4 | — | — | — | — | |
| Exact Search2024.12 | 58 | 45 | 20 | 6 | 16 | |
| Claude Code RawTime (s)=186, Cost ($)=2.222026.04 | 55.8 | — | — | — | — | |
| ReActLLM=GPT-4o mini2024.12 | 40 | 60 | 6 | 6 | 16 | |
| LLM-KBCILLM=GPT-4o mini2024.12 | 40 | 50 | 6 | 6 | 16 | |
| LLM-KBCI-RELLM=GPT-4o mini, Top-K Guess Reasoning=K=22024.12 | 40 | 50 | 6 | 6 | 16 | |
| MATMCD-RELLM=GPT-4o mini, Top-K Guess Reasoning=K=22024.12 | 40 | 50 | 6 | 6 | 16 | |
| Efficient-CDLMsLLM=GPT-4o mini2024.12 | 33 | 33 | 13 | 8 | 22 | |
| DirectLiNGAM2024.12 | 22 | 16 | 33 | 10 | 27 | |
| SCD-LLMLLM=GPT-4o mini2024.12 | 22 | 33 | 6 | 7 | 19 | |
| PC2024.12 | 15 | 14 | 20 | 9 | 25 | |
| MAC2024.12 | — | — | — | 6 | 19 |