1st Solution to QQ Browser 2021 AIAC Track 2

This repository is the winning solution to QQ Browser 2021 AI Algorithm Competition
Track 2 Automated Hyperparameter Optimization (CIKM AnalytiCup 2021).

Team: PKU-DAIR. Members: Huaijun Jiang, Yu Shen, Yang Li.

The solution is based on black-box optimization system OpenBox.


  • Technical Report on arXiv.
  • Presentation (Presented by Yu Shen): Slides, Video.

Repo structure

|--THPO_Kit_2021                # code for preliminary stage
|  |--pku_dair_openbox          # our solution
|  |--...
|--THPO_Final_Kit_2021          # code for final stage
|  |--pku_dair_openbox_final    # our solution
|  |--explore_data              # code for data exploration
|  |--...
|--docs                         # documents
|  |--...
|--README.md                    # this file
`--requirements.txt             # requirements

Note: We provide code for data exploration and documents in this repo.


  • python=3.6.8
conda create -n thpo python=3.6.8
  • openbox==0.7.16
pip install -r requirements.txt
  • Evaluation machine environment: linux 4.14, 64-bit operating system, memory 4G,
    CPU 2.3GHz dual-core Intel Core i5.


Preliminary Stage

cd THPO_Kit_2021

Local test:

./local_test.sh pku_dair_openbox


./prepare_submission.sh pku_dair_openbox

Final Stage

cd THPO_Final_Kit_2021

Local test:

./local_test.sh pku_dair_openbox_final


./prepare_submission.sh pku_dair_openbox_final

Please see README.md in inner folders for more details.


View Github