Welcome to the documentation for the Active Learning SEAMM plug-in#
Active Learning chooses what to label next when training a machine-learned force
field. A committee of models, each trained on a different part of the data, disagrees
where the model is likely to be wrong; the seamm-al command scores candidate
structures by that disagreement, selects the most useful ones within a computing
budget, trains the committees, checks that their disagreement tracks the real error,
and runs molecular dynamics with the committee to find new candidates. In SEAMM
flowcharts, the Active Learning step does the same with its sub-steps Train
Committee, Check Committee, Score Pool and Select Frames, and a trained committee is
a model chemistry of the xnn plug-in, for molecular dynamics that records where its
members disagree.