Skip to content

Mixed

Mixed data stream generator.

This generator is an implementation of a data stream with abrupt concept drift and boolean noise-free examples as described in 1.

It has four relevant attributes, two boolean attributes \(v, w\) and two numeric attributes \(x, y\) uniformly distributed from 0 to 1. The examples are labeled depending on the classification function chosen from below.

  • function 0: if \(v\) and \(w\) are true or \(v\) and \(z\) are true or \(w\) and \(z\) are true then 0 else 1, where \(z\) is \(y < 0.5 + 0.3 sin(3 \pi x)\)

  • function 1: The opposite of function 0.

Concept drift can be introduced by changing the classification function. This can be done manually or using ConceptDriftStream.

Parameters

  • classification_function

    Typeint

    Default0

    Which of the two classification functions to use for the generation. Valid options are 0 or 1.

  • seed

    Typeint | None

    DefaultNone

    Random seed for reproducibility.

  • balance_classes

    Typebool

    DefaultFalse

    Whether to balance classes or not. If balanced, the class distribution will converge to a uniform distribution.

Attributes

  • desc

    Return the description from the docstring.

Examples

from river.datasets import synth
dataset = synth.Mixed(seed = 42, classification_function=1, balance_classes = True)
for x, y in dataset.take(5):
    print(x, y)
{0: True, 1: False, 2: 0.2750, 3: 0.2232} 1
{0: False, 1: False, 2: 0.2186, 3: 0.5053} 0
{0: False, 1: True, 2: 0.8094, 3: 0.0064} 1
{0: False, 1: False, 2: 0.1010, 3: 0.2779} 0
{0: True, 1: False, 2: 0.37018, 3: 0.2095} 1

Methods

generate_drift

Generate drift by switching the classification function.

take

Iterate over the k samples.

Parameters

  • k'int'

Notes

The sample generation works as follows: The two numeric attributes are generated with the random generator initialized with the seed passed by the user (optional). The boolean attributes are either 0 or 1 based on the comparison of the random number generator and 0.5, the classification function decides whether to classify the instance as class 0 or class 1. The next step is to verify if the classes should be balanced, and if so, balance the classes.

The generated sample will have 4 relevant features and 1 label (it is a binary-classification task).


  1. Gama, Joao, et al. "Learning with drift detection." Advances in artificial intelligence-SBIA 2004. Springer Berlin Heidelberg, 2004. 286-295"