This repo implements incremental learning with scikit-learn by using a sample dataset containing a concept drift, i.e. a target variable that changes its meaning over time.
Users choose items they like. While they reject everything in the beginning f(g)=0,
a regular pattern emerges over time f(g)=-x.
I used this sample to understand the effects of partial_fit
with respect to both fitting time and quality of the training.
partial_all: partial_fit on all data until this pointpartial_step: partial_fit on batch datafull_step: regular fit on batch data (ignores all previous samples!)full_all: regular fit on all data until this point
PS: Please be aware that I have no idea what I'm doing.