TIME Benchmark Leaderboard

TIME Benchmark Leaderboard

A leaderboard for the TIME benchmark of probabilistic time series forecasting, with overall scores, per-dataset breakdowns and test window plots. Suits forecasting researchers and ML engineers comparing models.

Made by Real-TSF on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space is the leaderboard for TIME, a benchmark for probabilistic time series forecasting that scores models at several levels of detail. It reads its results from datasets on the Hugging Face Hub and shows them in a web page, so you can compare forecasting models without running them yourself.

Who it suits: It suits researchers and ML engineers who build or choose time series forecasting models and want a shared yardstick, and students learning how forecasting models are compared.

Industry | Who uses it | How they use it | Example
Research | Forecasting researchers | Compare their model with others across datasets | Check overall scores for a model family
Energy and utilities | Data scientists | Look at results on solar and similar series | Open the results for a solar dataset
Public health | Analysts | See how models did on case-count series | Open the results for a COVID dataset
Machine learning | ML engineers | Pick a forecasting model by looking at test windows | Study a prediction plot for one window

You can skip it if you need a forecasting tool for your own data, since this page only reports benchmark results.

Where to use it:

• A web browser: open the Space page on Hugging Face
• Windows, macOS or Linux: run a copy in Docker
• A local Python run, if you point it at local data

Use case | Who | What they get back
Compare models overall | Researchers | Scores aggregated across all datasets and horizons
Inspect one dataset | Analysts | A breakdown of performance for that dataset
Inspect a test window | ML engineers | A visual of the prediction for one test window

How to set it up: The online page needs nothing. A local copy needs Docker, or Python with the packages in requirements.txt.

• A web browser for the online demo
• Docker for the container route
• A Hugging Face token only if the data lives in a private dataset

1. To use it online, open the Space page on Hugging Face.
2. To run a copy, install the Hugging Face hub package.
3. Download the Space into a local folder.
4. Move into the folder and build the Docker image.
5. Run the container with the port published.

pip install -U huggingface_hub
hf download Real-TSF/TIME-leaderboard --repo-type space --local-dir TIME-leaderboard
cd TIME-leaderboard
docker build -t time-leaderboard .
docker run -it -p 7860:7860 time-leaderboard

The README also shows a plain Python install, and a way to read local data instead of the Hub.

pip install -r requirements.txt
python app.py

export USE_HF_HUB=false
python app.py

How to use it: Open the page and move between its three views. Start with the overall view, then open the dataset view for one dataset, then look at a single test window with its prediction plot.

Tool | What it does
Overall performance | Shows metrics aggregated across all datasets and horizons
Dataset-level analysis | Breaks down performance by individual dataset
Window-level visualization | Shows a detailed test window with its prediction plot

Example: You open the overall view to see how models rank, then pick one dataset where your model of interest did well. Last, you open one test window to see how its forecast lined up with the real values.

For newcomers and experts: If you are new to this, forecasting means predicting future values of a series, such as daily case counts. Start with the overall view to see which models do well, then open one dataset.

If you know the field, the benchmark covers probabilistic forecasts at several granularities. The app reads datasets and model outputs from a Hub dataset, controlled by environment variables for the token, repository, hub switch and cache folder. The expected layout holds the series data, per-model result folders and a dataset config file.

The app reads data from the Hub or local files
Results are grouped by model and dataset
You open the overall, dataset or window view
Scores and prediction plots appear

Good to know: The release is for research only, in support of an academic paper, and is not evaluated for every downstream use. The authors advise checking accuracy, safety and fairness before relying on any model it ranks.

• License: apache-2
• Cost to run: none named, but a Hugging Face token is needed for private datasets
• Data it reaches: benchmark datasets and results on the Hugging Face Hub, or local files if you switch the hub off
• Not for: production use without your own checks, since it is a research release

On Hugging Face, 13 November 2026

• Likes: 32
• Built with: docker
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/Real-TSF/TIME-leaderboard
• Files: https://huggingface.co/spaces/Real-TSF/TIME-leaderboard/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.