Machine learning methods are often seen as black boxes which are difficult to understand. Automated machine learning frameworks tend to take these boxes, put them into a dark room, lock the door, and run away (watch out for sarcasm!). So instead of improving interpretability directly, let's conduct a benchmark in a meaningful way in order to learn more about them.
There have been some approaches in this direction already. See, e.g., the AutoML Benchmark, the work of STATWORX, and the benchmark accompanying the Auto-Sklearn publication. However, they are all lacking with regards to the underlying pipeline system using to setup the benchmark, as well as the exploration capabilities of the results.
Here, we build a sustainable (i.e. reproducible, adaptable, and transparent) workflow to automatically benchmark a multitude of models against a diverse set of data. The models are existing auto-ML frameworks which each have their own advatanges and disdvantages related to ease-of-use, execution speed, and predictive performance. All of these features will become apparent in this benchmark. The datasets try to be as representative as possible and cover a wide range of applications. They thus serve as a reasonable playground for the aforementioned models. Finally, the results are displayed in an interactive dashboard which allows an in-depth exploration of the generated performance evaluation.
- XGBoost: optimized distributed gradient boosting (this will serve as baseline to calibrate the results)
- auto-sklearn: "extend Auto-sklearn with a new, simpler meta-learning technique"
- PyCaret: "end-to-end machine learning and model management tool"
- TPOT: "optimizes machine learning pipelines using genetic programming"
test_dataset: just a dummy dataset to make sure everything works
Execute the following command to run all models on all datasets:
$ snakemake -j 1 -pr --use-condaAfterwards, execute python results/dashboard.py to enter an interactive dashboard for exploring the results.
To add a new dataset, you simply need to add a single CSV file to ./resources/datasets/{dataset}.csv.
Each response variable/column needs to be prefixed with target__ (and there has to be exactly one). All other columns are treated as covariates.
A new model can be added by implementing its training procedure and putting the script into ./resources/models/{model}.py. Each script consists of a single main function which takes X_train and y_train as input.