search

EpistasisLab / tpot

A Python Automated Machine Learning tool that optimizes machine learning pipelines using genetic programming.
http://epistasislab.github.io/tpot/
GNU Lesser General Public License v3.0
9.69k stars 1.57k forks source link

How TPOT tune the parameters? #743

Closed shuaianuoe closed 5 years ago

shuaianuoe commented 6 years ago

Hi all,

I am interested in how TPOT tune the parameters.

As a beginner, i know that TPOT is based on genetic programming. I have a question, does the genetic algorithm has the ability to tune the parameters, such as the lightgbm's paremeters? In other words, what strategy does TPOT use to tune parameters? is grid search or Bayesian optimization or what other methods?

Thank you very much and best regards! shuai

weixuanfu commented 6 years ago

TPOT just uses genetic programming to tune the parameters. If the lightgbm has a scikit-learn API then the answer is yes.

A related issue #335 #688

weixuanfu commented 6 years ago

Here is lightgbm's scikit-learn API. Please check the link for more details about how to make a custom configuration for the operators and parameters in TPOT.