Updated on 2023-12-13 GMT+08:00

TPE Algorithm

The tree-structured parzen estimator (TPE) algorithm uses the Gaussian mixture model to learn the model hyperparameters. On each trial, for each parameter, TPE fits one Gaussian mixture model l(x) to the set of parameter values associated with the best objective values, and another Gaussian mixture model g(x) to the remaining parameter values. It chooses the hyperparameter value that maximizes the ratio l(x)/g(x).

Table 1 TPE algorithm parameters

Parameter

Description

Recommended Value

num_samples

Number of times to sample from the hyperparameter space

The value is an integer ranging from 10 to 20. The larger the value, the longer the search time and the better the effect.

n_initial_points

Number of random evaluations of the objective function before starting to approximate it with tree parzen estimators

The value is an integer. You are advised not to change it.

gamma

Quantile of the TPE algorithm to divide l(x) and g(x)

This value is a float ranging from 0 to 1. You are advised not to change it.