Skip to content


Scales data using exponentially weighted moving average and variance.

Under the hood, a exponentially weighted running mean and variance are maintained for each feature. This can potentially provide better results for drifting data in comparison to preprocessing.StandardScaler. Indeed, the latter computes a global mean and variance for each feature, whereas this scaler weights data in proportion to their recency.


  • alpha – defaults to 0.3

    This parameter is passed to stats.EWVar. It is expected to be in [0, 1]. More weight is assigned to recent samples the closer alpha is to 1.


Consider the following series which contains a positive trend.

>>> import random

>>> random.seed(42)
>>> X = [
...     {'x': random.uniform(4 + i, 6 + i)}
...     for i in range(8)
... ]
>>> for x in X:
...     print(x)
{'x': 5.278}
{'x': 5.050}
{'x': 6.550}
{'x': 7.446}
{'x': 9.472}
{'x': 10.353}
{'x': 11.784}
{'x': 11.173}

This scaler works well with this kind of data because it uses statistics that assign higher weight to more recent data.

>>> from river import preprocessing

>>> scaler = preprocessing.AdaptiveStandardScaler(alpha=.6)

>>> for x in X:
...     print(scaler.learn_one(x).transform_one(x))
{'x': 0.0}
{'x': -0.816}
{'x': 0.812}
{'x': 0.695}
{'x': 0.754}
{'x': 0.598}
{'x': 0.651}
{'x': 0.124}



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)


Transformer: self


Transform a set of features x.


  • x (dict)


dict: The transformed values.