What is this?
Nimbus turns any point forecast into a probabilistic forecast
— showing a range of possible outcomes and their likelihood rather than a
single predicted value. It fits a Gaussian process model to your data and
displays fan-chart confidence bands.
How to use
- Example data — select a built-in dataset from the dropdown
to load sample forecasts (sine waves, ERCOT load, solar, wind, LMPs).
- Paste your own data — type or paste up to 30 rows into
the data grid. Use multiple columns for separate observations at each time
step.
- Create forecast — click the button to fit a Gaussian
process model. The chart shows the mean prediction with gold confidence
bands.
- Adjust — after fitting, use the Shape, Uncertainty,
Structure, and Distribution controls to fine-tune the model without
re-running the full search.
- Quantile table — after fitting, a table of quantile
values appears below the chart.
Key concepts
- Gaussian Process
- A non-parametric model that defines a distribution over functions.
Given observed data, it produces a posterior mean and variance at every
point — yielding natural confidence bands.
- Kernel search
- Nimbus automatically searches over kernel compositions (RBF, periodic,
white noise) to find the best structure for your data, inspired by the
Automatic Statistician.
- Fan chart
- The gold shaded bands show prediction intervals at increasing
confidence levels, from the median outward.
References
- Lloyd, J.R., Duvenaud, D., Grosse, R., Tenenbaum, J.B., & Ghahramani, Z. (2014).
Automatic Construction and Natural-Language Description of Nonparametric Regression Models.
arXiv:1402.4304.
arxiv.org/abs/1402.4304
Disclaimer & Terms of Use
This tool is provided free of charge by Distill Energy for informational
and educational purposes. You are free to use and reproduce this tool and
its outputs. However, Distill Energy provides this tool "as is" without
warranty of any kind, express or implied, and disclaims all liability for
any damages, losses, or decisions arising from its use or its results.
The probabilistic forecasts generated are statistical estimates and should
not be relied upon as the sole basis for financial, operational, or
investment decisions.
