DynamicQuantizer

Adapted version of the Quantizer Observer (QO)¹ that is applied to Stochastic Gradient Trees (SGT).

This feature quantizer starts by partitioning the inputs using the passed radius value. As more splits are created in the SGTs, new feature quantizers will use std * std_prop as the quantization radius. In the expression, std represents the standard deviation of the input data, which is calculated incrementally.

Parameters

• radius (float) – defaults to 0.5

  The initial quantization radius.

• std_prop (float) – defaults to 0.25

  The proportion of the standard deviation that is going to be used to define the radius value for new quantizer instances following the initial one.

Methods

update

References

---

1. Mastelini, S.M. and de Leon Ferreira, A.C.P., 2021. Using dynamical quantization to perform split attempts in online tree regressors. Pattern Recognition Letters. ↩