benchm-ml

Why we included this project

Choosing between the common open source machine learning libraries is usually done on hearsay, and this repository exists to replace that with measurements. It runs a small set of binary classification algorithms, from logistic regression to random forests, gradient boosting, and deep networks, through implementations such as scikit-learn, R packages, xgboost, lightgbm, H2O, Vowpal Wabbit, and Spark MLlib, then records training time, memory use, and AUC as the dataset grows from 10,000 to 10 million rows. The workloads mirror a typical business problem like credit scoring or churn prediction, with numeric and categorical inputs and a few thousand features, so the results speak to real tabular data rather than toy examples. The author is unusually transparent about the single node hardware, the exact tool versions, and what the study does not cover, which makes it easier to judge whether the numbers carry over to your own setup. It also dates itself honestly, and points to the author's successor benchmarks when you need more recent gradient boosting results.