TIME Benchmark Leaderboard

A unified benchmark for time series probabilistic forecasting with multiple granularity evaluation.

Features

• Overall Performance: Aggregated metrics across all datasets and horizons
• Dataset-level Analysis: Performance breakdown by individual datasets
• Window-level Visualization: Detailed test window analysis with prediction visualization

Configuration

Environment Variables

The app reads data from HuggingFace Hub. Configure the following environment variables:

For HuggingFace Space Deployment

快速部署（推荐）

# 1. 复制 timebench 模块到 leaderboard_app
cd /home/eee/qzz/TIME
cp -r src/timebench leaderboard_app/

# 2. 进入 leaderboard_app 目录
cd leaderboard_app

# 3. 运行部署脚本
chmod +x deploy.sh
./deploy.sh YOUR_USERNAME YOUR_SPACE_NAME

手动部署

详细步骤请参考 DEPLOY.md

重要: 部署前需要：

1. 创建 HuggingFace Space: https://huggingface.co/new-space
2. 在 Space Settings → Repository secrets 中添加 HF_TOKEN
3. 确保数据已上传到 TIME-benchmark/TIME-1.0 Dataset

For Local Development

Set USE_HF_HUB=false to use local data:

export USE_HF_HUB=false
python app.py

Installation

pip install -r requirements.txt
python app.py

Data Structure

The app expects the following data structure in the HuggingFace Dataset:

HF_REPO/
├── data/
│   └── hf_dataset/           # Time series datasets
│       ├── ECDC_COVID/
│       ├── Australia_Solar/
│       └── ...
├── output/
│   └── results/              # Model evaluation results
│       ├── moirai_small/
│       ├── chronos_base/
│       └── ...
└── config/
    └── datasets.yaml         # Dataset configurations

Ethical Considerations

This release is for research purposes only in support of an academic paper. Our models, datasets, and code are not specifically designed or evaluated for all downstream purposes. We strongly recommend users evaluate and address potential concerns related to accuracy, safety, and fairness before deploying this model.