Adaptive Windowing method for concept drift detection.

ADWIN (ADaptive WINdowing) is a popular drift detection method with mathematical guarantees. ADWIN efficiently keeps a variable-length window of recent items; such that it holds that there has no been change in the data distribution. This window is further divided into two sub-windows $$(W_0, W_1)$$ used to determine if a change has happened. ADWIN compares the average of $$W_0$$ and $$W_1$$ to confirm that they correspond to the same distribution. Concept drift is detected if the distribution equality no longer holds. Upon detecting a drift, $$W_0$$ is replaced by $$W_1$$ and a new $$W_1$$ is initialized. ADWIN uses a significance value $$\delta=\in(0,1)$$ to determine if the two sub-windows correspond to the same distribution.

## Parameters¶

• delta

Default0.002

Significance value.

• clock

Default32

How often ADWIN should check for change. 1 means every new data point, default is 32. Higher values speed up processing, but may also lead to increased delay in change detection.

• max_buckets

Default5

The maximum number of buckets of each size that ADWIN should keep before merging buckets (default is 5).

• min_window_length

Default5

The minimum length of each subwindow (default is 5). Lower values may decrease delay in change detection but may also lead to more false positives.

• grace_period

Default10

ADWIN does not perform any change detection until at least this many data points have arrived (default is 10).

## Attributes¶

• drift_detected

Whether or not a drift is detected following the last update.

• estimation

Estimate of mean value in the window.

• n_detections

• total

• variance

• width

Window size

## Examples¶

import random
from river import drift

rng = random.Random(12345)

data_stream = rng.choices([0, 1], k=1000) + rng.choices(range(4, 8), k=1000)

for i, val in enumerate(data_stream):
print(f"Change detected at index {i}, input value: {val}")

Change detected at index 1023, input value: 4


## Methods¶

update

Update the change detector with a single data point.

Apart from adding the element value to the window, by inserting it in the correct bucket, it will also update the relevant statistics, in this case the total sum of all values, the window width and the total variance.

Parameters

• x'numbers.Number'

Returns

DriftDetector: self

1. Albert Bifet and Ricard Gavalda. "Learning from time-changing data with adaptive windowing." In Proceedings of the 2007 SIAM international conference on data mining, pp. 443-448. Society for Industrial and Applied Mathematics, 2007.