Skip to content


Jackknife method for regression.

This is a conformal prediction method for regression. It is based on the jackknife method. The idea is to compute the quantiles of the residuals of the regressor. The prediction interval is then computed as the prediction of the regressor plus the quantiles of the residuals.

This works naturally online, as the quantiles of the residuals are updated at each iteration. Each residual is produced before the regressor is updated, which ensures the predicted intervals are not optimistic.

Note that the produced intervals are marginal and not conditional. This means that the intervals are not adjusted for the features x. This is a limitation of the jackknife method. However, the jackknife method is very simple and efficient. It is also very robust to outliers.


  • regressor (base.Regressor)

    The regressor to be wrapped.

  • confidence_level (float) – defaults to 0.95

    The confidence level of the prediction intervals.

  • window_size (int) – defaults to None

    The size of the window used to compute the quantiles of the residuals. If None, the quantiles are computed over the whole history. It is advised to set this if you expect the model's performance to change over time.


>>> from river import conf
>>> from river import datasets
>>> from river import linear_model
>>> from river import metrics
>>> from river import preprocessing
>>> from river import stats

>>> dataset = datasets.TrumpApproval()

>>> model = conf.RegressionJackknife(
...     (
...         preprocessing.StandardScaler() |
...         linear_model.LinearRegression(intercept_lr=.1)
...     ),
...     confidence_level=0.9
... )

>>> validity = stats.Mean()
>>> efficiency = stats.Mean()

>>> for x, y in dataset:
...     interval = model.predict_one(x, with_interval=True)
...     validity = validity.update(y in interval)
...     efficiency = efficiency.update(interval.width)
...     model = model.learn_one(x, y)

The interval's validity is the proportion of times the true value is within the interval. We specified a confidence level of 90%, so we expect the validity to be around 90%.

>>> validity
Mean: 0.903097

The interval's efficiency is the average width of the intervals.

>>> efficiency
Mean: 3.593173

Lowering the confidence lowering will mechanically improve the efficiency.



Fits to a set of features x and a real-valued target y.


  • x
  • y
  • kwargs




Predict the output of features x.


  • x
  • with_interval – defaults to False
  • kwargs


The prediction.