{"data":{"kind":"file","path":"README.md","version_id":"ftk3gy772ituivlp8dhsf5nn","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1057,"modified_at":"2026-08-12T10:25:25.363000","content_hash":"6e3c3814fb771dbf7bb6c2a989be9e9216392fd01055189f9ca66e117cb9304a"},"entries":[],"content":"# Time-Series Forecasting Environment\n\nEvaluates an LLM's ability to forecast time-series data, detect underlying patterns (trend, seasonality, noise), and explain forecasting methodology.\n\n## Tasks\n\n1. **Next-value prediction**: Given a sequence of numeric values, predict the next N values.\n2. **Pattern detection**: Identify trend direction, seasonality period, and noise level.\n3. **Methodology explanation**: Explain the reasoning behind the forecast.\n\n## Reward Functions\n\n- `forecast_accuracy`: MAPE-based accuracy of predicted values vs ground truth\n- `pattern_detection`: Correctly identifies trend, seasonality, and noise characteristics\n- `explanation_quality`: Assesses depth and correctness of methodology explanation\n\n## Dataset\n\nSynthetic time-series with configurable:\n- Trend (linear, exponential, logarithmic)\n- Seasonality (daily, weekly, monthly periods)\n- Noise levels (low, medium, high)\n- Sequence lengths (20-100 points)\n\n## Usage\n\n```python\nimport verifiers as vf\nenv = vf.load_environment(\"alexnguyen/time-series-forecasting\")\n```\n","encoding":"utf-8","truncated":false,"total_bytes":1057},"status":null}