Contextual forecasting on Context Is Key
54.5SMAPECOUNTS
Evaluation results
| Method | Links | ||
|---|---|---|---|
| COUNTSTraining Stage=Reinforcement Learning (RL)2025.10 | 54.5 | 58.1 | |
| COUNTSTraining Stage=SFT, Fine-tuning Protocol=fine-tuned on the CiK dataset2025.10 | 61.7 | 64.8 | |
| TsLLM2025.10 | 64.5 | 64.7 | |
| ChatTs2025.10 | 68.4 | 69.2 | |
| ChatTs2025.10 | 68.4 | 69.2 | |
| COUNTSTraining Stage=Pre-RL2025.10 | 68.7 | 70.1 | |
| ChatTime2025.10 | 70.1 | 71.4 | |
| ChatTime2025.10 | 70.1 | 71.4 | |
| o4-mini2025.10 | 72.6 | 70.5 | |
| o4-mini2025.10 | 72.6 | 70.5 | |
| Linear Regression2025.10 | 75.4 | 101.5 | |
| Linear Regression2025.10 | 75.4 | 101.5 | |
| XGBoost2025.10 | 76.8 | 80.2 | |
| XGBoost2025.10 | 76.8 | 80.2 | |
| ARIMA2025.10 | 90.7 | 134.4 | |
| ARIMA2025.10 | 90.7 | 134.4 | |
| Gemini 2.5 Pro2025.10 | 90.8 | 98.1 | |
| Gemini 2.5 Pro2025.10 | 90.8 | 98.1 | |
| Qwen 2.5Parameters=7B2025.10 | 92.6 | 139.2 | |
| ETS2025.10 | 110 | 204.2 |