NEWS.md

In DataSystemsGroupUT/SmartML: Machine Learning Automation

SmartML 0.3.0.1

• Hotfix, fixed some dependency issues relating to dplyr

SmartML 0.3.0

Features

• Added Ranger, XGBoost, fastNaiveBayes and LiblineaR high performing algorithms
• Added the autoRLearn_ function, which assumes that the data is in perfect shape and can be loaded from a dataframe, unlike autoRLearn which can only load from a data file outside R.
• Added Hyperband and Bayesian Optimization Hyperband to the new autoRLearn_
• Added some extra temporary dependencies which will be removed in the following months (all tidyverse packages other than purrr)
• Fixed some small mistakes in the code and jsons

Current Roadmap

• fix metalearning, at the moment it doesn't work. There's something wrong with the AWS server we are using.
• change the dplyr back end to use data.table with dtplyr
• merge autoRLearn and autoRLearn_ into a single function, which can both load from a data file and in R.
• Rewrite SMAC, as requested by Sherif.

Extra info

• brurucy is a new and active maintainer
• Nightly and experimental versions, independent from the Data Systems Lab, are being developed at https://github.com/brurucy/witchcraft
• Updates will be conservative and focused on non-breaking changes, up to release 1.0.

DataSystemsGroupUT/SmartML documentation built on Nov. 24, 2020, 1:31 p.m.