auto-flow / ml-pipeline-experiment

Machine-Learning Pipeline Experiment

Experiments about several Bayesian Optimization Algorithms performance in Machine-Learning Pipeline Space

In this Experiments, our's ETPE(Embedding Tree Structure Estimators) implemented in UltraOpt is out perform other TPE algorithms, for example, initial TPE implemented in HyperOpt .

Build Machine-Learning Pipeline Space

We provide several scripts to build Machine-Learning Pipeline Space. We have built three Space which is public available:

• 146594 dataset
• 189863 dataset
• 189864 dataset

Start-Up Experiments between Bayesian Algorithms

After clone this repositary and install necessary requirements, you can Start-Up Experiments between Bayesian Algorithms in following operations:

1. clone this repositary and install necessary requirements

git clone https://github.com/auto-flow/ml-pipeline-experiment
cd ml-pipeline-experiment
pip install -r requirements.txt

2. Start-Up Experiments between Bayesian Algorithms

cat experiments/run_experiments_146594.sh|parallel -j 6
cat experiments/run_experiments_189863.sh|parallel -j 6
cat experiments/run_experiments_189864.sh|parallel -j 6

Results and Conclusion

In this Experiments, our's ETPE(Embedding Tree Structure Estimators) implemented in UltraOpt is out perform other TPE algorithms.