I am starting a new project on a tight budget and cannot afford expensive enterprise software. What are the most reliable open-source Python libraries for automated machine learning? I am specifically looking for something that handles preprocessing and model selection well for a medium-sized tabular dataset.
3 answers
If you are looking for ease of use, PyCaret is arguably the best library for automated machine learning right now. It is a low-code library that wraps around Scikit-Learn, XGBoost, and LightGBM. With just a few lines of code, you can compare multiple models, tune hyperparameters, and even plot your results. If you want something more mathematically rigorous that uses genetic programming, check out TPOT. It takes longer to run because it explores a vast search space of pipelines, but it often finds very creative feature combinations that a human might never think to try.
Does your dataset have a lot of missing values or categorical variables that require complex encoding before running automated machine learning?
Auto-Sklearn is fantastic because it uses Bayesian optimization and ensemble building, making it a very robust choice for automated machine learning on tabular data.
I’ve had great luck with Auto-Sklearn on Linux. It’s a bit of a pain to install on Windows, but the performance is top-tier for open-source.
Yes, it’s quite messy. I have about 40% missing data in three key columns and a lot of high-cardinality categorical features that usually break standard encoders.