Zero-Shot Time Series Forecasting with TabPFN
- 12-05-2026: 🚀 v1.1.0 — defaults now ship the TabPFN-TS-3 finetuned checkpoint from the TabPFN-3 report for SOTA forecasting out of the box. Requires
tabpfn>=8.0.0. - 27-05-2025: 📝 New paper version and v1.0.0 release! Strong GIFT-EVAL results, new AutoSeasonalFeatures, improved CalendarFeatures.
- 27-01-2025: 🚀 Ranked 1st on GIFT-EVAL benchmark[1]!
- 10-10-2024: 🚀 TabPFN-TS paper accepted to NeurIPS 2024 TRL and TSALM workshops!
[1] Last checked on: 10/03/2025
We demonstrate that the tabular foundation model TabPFN, combined with lightweight feature engineering, enables zero-shot time series forecasting for both point and probabilistic tasks. On the GIFT-EVAL benchmark, our method achieves performance on par with top-tier models across both evaluation metrics.
As of v1.1.0 the package ships with the finetuned TabPFN-TS-3 checkpoint by default — a tabular foundation model that, despite being pretrained purely on synthetic data, achieves SOTA results on fev-bench. See the TabPFN-3 report for benchmarks.
Our work proposes to frame univariate time series forecasting as a tabular regression problem.
Concretely, we:
- Transform a time series into a table
- Extract features and add them to the table
- Perform regression on the table using TabPFN
- Use regression results as time series forecasting outputs
For more details, see our paper and the TabPFN-3 report.
TabPFN-TS treats forecasting as univariate, target-history + known-future covariate regression. Concretely:
| Input column type | Used? | Notes |
|---|---|---|
| Target (univariate or multivariate) | ✅ | Multivariate targets are decomposed into N independent univariate forecasts. |
| Known dynamic covariates (e.g. holidays) | ✅ | Future values must be supplied in future_df. |
| Past dynamic covariates (e.g. weather observations) | ❌ | Dropped — no future values available. |
| Static covariates (e.g. item_id, store_size) | ❌ | Currently dropped. |
| Engineered temporal features (calendar, sin/cos, …) | ✅ | Auto-generated by TABPFN_TS_DEFAULT_FEATURES. |
If your task relies on past_dynamic or static_columns, performance may be lower than dedicated multivariate / static-aware forecasters.
- Zero-shot forecasting: this method is extremely fast and requires no training, making it highly accessible for experimenting with your own problems.
- Point and probabilistic forecasting: it provides accurate point forecasts as well as probabilistic forecasts.
- Known-future covariate support: pass covariates whose future values are known at forecast time (calendar, holidays, scheduled events, demand or weather forecasts) directly alongside the target — see Covariate model for details on what's supported.
On top of that, thanks to tabpfn-client from Prior Labs, you won't even need your own GPU to run fast inference with TabPFN. 😉 We have included tabpfn-client as the default engine in our implementation.
You can install the package via pip:
pip install tabpfn-time-seriesv1.1.0+ ships the TabPFN-TS-3 finetuned checkpoint by default. In LOCAL mode tabpfn downloads it automatically on first init — accept the license at ux.priorlabs.ai first.
To install the package in editable mode with all development dependencies, run the following command in your terminal:
pip install -e ".[dev]"
# or with uv
uv pip install -e ".[dev]"from tabpfn_time_series import TabPFNTSPipeline
pipeline = TabPFNTSPipeline() # uses the cloud client by default — no GPU needed
predictions = pipeline.predict_df(context_df, prediction_length=24)For full walkthroughs:
| I want to... | Notebook |
|---|---|
| Use it on my project | quickstart.ipynb |
| Understand how it works | how-it-works.ipynb |
More examples live in the examples/ directory.
If TabPFN-TS is useful in your research, please cite the paper (arXiv:2501.02945) and the TabPFN-3 report for the v3 checkpoint.
This project collects anonymous usage telemetry by default.
The data is used exclusively to help us understand how the library is being used and to guide future improvements.
- No personal data is collected
- No code, model inputs, or outputs are ever sent
- Data is strictly anonymous and cannot be linked to individuals
We only collect high-level, non-identifying information such as:
- Package version
- Python version
- How often fit and inference are called, including simple metadata like the dimensionality of the input and the type of task (e.g., classification vs. regression) (:warning: never the data itself)
See the Telemetry documentation for the full details of events and metadata.
This data is processed in compliance with the General Data Protection Regulation (GDPR) principles of data minimization and purpose limitation.
For more details, please see our Privacy Policy.
If you prefer not to send telemetry, you can disable it by setting the following environment variable:
export TABPFN_DISABLE_TELEMETRY=1