Compute the adaptive forgetting factor - no change detection

Description

Original implementation in R of the AFF

Usage

1
AFF_scaled_stream_no_change_detection(stream, BL, params)

Arguments

stream

The stream of observations.

BL

The burn-in length, used to estimate the mean and variance.

params

An unnamed list of parameters for the FFF algorithm. Consists of:

lambda

The value of the fixed forgetting factor (FFF). Should be in the range [0,1].

p

The value of the significance threshold, which was later renamed alpha (in the paper, not in this function).

resettozero

A flag; if it zero, then the ffmean will be reset to zero after each change. Usually set to 1 (i.e. do not reset).

u_init

The initial value of u. Should be set to 0.

v_init

The initial value of v. Should be set to 0.

w_init

The initial value of w. Should be set to 0.

affmean_init

The initial value of the forgetting factor mean, ffmean. Should be set to 0.

affvar_init

The initial value of the forgetting factor variance, ffvar. Should be set to 0.

low_bound

The lower bound for lambda. Usually set to 0.6.

up_bound

The upper bound for lambda. Usually set to 1.

signchosen

The sign used in the gradient. descent. Usually set to -1.

alpha

The value of the step size in the gradient descent step. In the paper it is referred to as ε. Usually 0.01, or otherwise 0.1 or 0.001.

Value

A vector with the values of the adaptive forgetting factor \overrightarrow{λ}.

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.