Show HN: Compare AutoML frameworks on 10 Tabular Kaggle competitions

I'm working on an AutoML system for tabular datasets. It is called MLJAR and is available as open-source with code on GitHub: https://github.com/mljar/mljar-supervised

I've compared my AutoML with other systems on 10 tabular datasets from Kaggle. The final result is the Percentile Rank in the Private Leaderboard (evaluated by Kaggle). The results other than MLJAR systems are from AutoGluon paper.

    Dataset      Auto-WEKA   auto-sklearn   TPOT    H2O AutoML   GCP-Tables   AutoGluon   MLJAR  
 -------------- ----------- -------------- ------- ------------ ------------ ----------- ------- 
  ieee-fraud         0.119          0.349                             0.119       0.322   0.172  
  value              0.114          0.319   0.325        0.377                    0.415   0.445  
  walmart                           0.390   0.379                     0.398       0.384   0.423  
  transaction        0.131          0.329   0.326                     0.404       0.406   0.463  
  porto              0.158          0.331   0.315        0.406        0.434       0.462   0.540  
  allstate           0.124          0.310   0.237        0.352         0.74       0.706   0.764  
  mercedes           0.160          0.444   0.547        0.363        0.658       0.169   0.879  
  otto               0.145          0.717   0.597        0.729        0.821       0.988   0.924  
  satisfaction       0.235          0.408   0.495        0.740        0.763       0.823   0.975  
  bnp-paribas        0.193          0.412   0.460        0.417        0.440       0.986   0.986  

The higher the value, the better. The 1st place solution in the Kaggle competition will get Percentile Rank equal 1.0. You can see that some AutoML frameworks jump into the Top-10% of the competition (without any human help)!

I think that my AutoML system is quite advanced:

- it can generate new features with K-Means or Golden Features Search

- it has many ML algorithms available, can tune them and train (with early stopping if applicable), in selected time regime,

- can stack models in complex ensembles

- creates interpretations for ML models: SHAP plots, permutation-based importance, decision tree visualizations ...

- automatically generates documentation to Markdown or HTML (works like a dream in Jupyter notebook)

I hope that many data scientists will benefit from my AutoML system. I put a lot of effort into it.