Sine

Sine generator.

This generator is an implementation of the dara stream with abrupt concept drift, as described in Gama, Joao, et al. ¹.

It generates up to 4 relevant numerical features, that vary from 0 to 1, where only 2 of them are relevant to the classification task and the other 2 are optionally added by as noise. A classification function is chosen among four options:

1. SINE1. Abrupt concept drift, noise-free examples. It has two relevant attributes. Each attributes has values uniformly distributed in [0, 1]. In the first context all points below the curve \(y = sin(x)\) are classified as positive.

2. Reversed SINE1. The reversed classification of SINE1.

3. SINE2. The same two relevant attributes. The classification function is \(y < 0.5 + 0.3 sin(3 \pi x)\).

4. Reversed SINE2. The reversed classification of SINE2.

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

Two important features are the possibility to balance classes, which means the class distribution will tend to a uniform one, and the possibility to add noise, which will, add two non relevant attributes.

Parameters

• classification_function

  Type → int

  Default → 0

  Classification functions to use. From 0 to 3.

• seed

  Type → int | None

  Default → None

  Random seed for reproducibility.

• balance_classes

  Type → bool

  Default → False

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

• has_noise

  Type → bool

  Default → False

  Adds 2 non relevant features to the stream.

Attributes

• desc

  Return the description from the docstring.

Examples

from river.datasets import synth

dataset = synth.Sine(classification_function = 2, seed = 112,
                     balance_classes = False, has_noise = True)

for x, y in dataset.take(5):
    print(x, y)

{0: 0.4812, 1: 0.6660, 2: 0.6198, 3: 0.6994} 1
{0: 0.9022, 1: 0.7518, 2: 0.1625, 3: 0.2209} 0
{0: 0.4547, 1: 0.3901, 2: 0.9629, 3: 0.7287} 0
{0: 0.4683, 1: 0.3515, 2: 0.2273, 3: 0.6027} 0
{0: 0.9238, 1: 0.1673, 2: 0.4522, 3: 0.3447} 0

Methods

generate_drift

Generate drift by switching the classification function at random.

take

Iterate over the k samples.

Parameters

• k — 'int'

Notes

The sample generation works as follows: The two attributes are generated with the random number generator. The classification function defines whether to classify the instance as class 0 or class 1. Finally, data is balanced and noise is added, if these options are set by the user.

The generated sample will have 2 relevant features, and an additional two noise features if has_noise is set.

---

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