yhatlabs/ChakraTS
Time Series Forecasting • Updated
AI & ML interests
Building foundation models for time series and tabular data, and systems that forecast real-world events.
Recent Activity
Organization Card
YHat Labs builds foundation models for time series and tables. ChakraTS forecasts any time series and ChakraTab predicts any column of a table: pretrained, one API call, no training or tuning. This organization holds the model cards; results files and reproduction code are on GitHub.
- 🌐 Models and results
- 📖 Docs
- 🔑 Get API access
- 📊 Benchmark results and reproduction notebooks
- ✨ GitHub
The Chakra models
| Model | Data | What it does | Availability |
|---|---|---|---|
| ChakraTS | Time series | Zero-shot probabilistic forecasts at any quantile level, known-future and past covariates, several targets at once, any regular frequency | API, commercial use |
| ChakraTab | Tables | Classification with calibrated probabilities, regression with intervals and full predictive distributions, fit once and predict many times | API, commercial use |
| ChakraTS-Lab | Time series | Research configuration of ChakraTS, used for benchmark entries | Evaluation on request |
Quick start
pip install yhatlabs
import yhatlabs
client = yhatlabs.Client() # reads YHAT_API_KEY
forecast = client.forecast(df, h=28) # a long table: id, time, value
preds = client.predict(train, "y", test) # any table, predicting column y
Benchmarks
We evaluate on the public leaderboards under their official protocols and publish the result files and the notebooks that regenerate them. Standing at the time of submission:
| Benchmark | Model | Standing |
|---|---|---|
| fev-bench | ChakraTS | 1st by win rate, 2nd by skill score |
| TIME | ChakraTS | 2nd of 31 |
| GIFT-Eval | ChakraTS | 3rd among non-agentic models |
| TabArena | ChakraTab | Top 5 (under review) |
Benchmark data is never used for anything other than scoring.
Contact
datasets 0
None public yet