Skip to content


Randomly selects features with an inclusion trial.

When a new feature is encountered, it is selected with probability p. The number of times a feature needs to beseen before it is added to the model follows a geometric distribution with expected value 1 / p. This feature selection method is meant to be used when you have a very large amount of sparse features.


  • p (float)

    Probability of including a feature the first time it is encountered.

  • seed (int) – defaults to None

    Random seed value used for reproducibility.


>>> from river import datasets
>>> from river import feature_selection
>>> from river import stream

>>> selector = feature_selection.PoissonInclusion(p=0.1, seed=42)

>>> dataset = iter(datasets.TrumpApproval())

>>> feature_names = next(dataset)[0].keys()
>>> n = 0

>>> while True:
...     x, y = next(dataset)
...     xt = selector.transform_one(x)
...     if xt.keys() == feature_names:
...         break
...     n += 1

>>> n



Return a fresh estimator with the same parameters.

The clone has the same parameters but has not been updated with any data. This works by looking at the parameters from the class signature. Each parameter is either - recursively cloned if it's a River classes. - deep-copied via copy.deepcopy if not. If the calling object is stochastic (i.e. it accepts a seed parameter) and has not been seeded, then the clone will not be idempotent. Indeed, this method's purpose if simply to return a new instance with the same input parameters.


Update with a set of features x.

A lot of transformers don't actually have to do anything during the learn_one step because they are stateless. For this reason the default behavior of this function is to do nothing. Transformers that however do something during the learn_one can override this method.


  • x (dict)
  • kwargs


Transformer: self


Transform a set of features x.


  • x (dict)


dict: The transformed values.