Shifu 0.2.4 Work Flow Change
new -> init -> stats -> varselect -> normalize -> train -> postttrain -> eval
new -> init -> stats -> normalize -> varselect -> train -> postttrain -> eval
In real scenarios, modelers select variables and then train models to see what kind of composition of variables can lead to good performance. Our old process is to select variables firstly and then only normalize selected variables and train models. Every time modelers need run 'normalize' after variable re-selection.
To solve such issue, all variables will be normalized firstly, then no matter what kind of variables are selected, no need to do 'normalize' step again.
If do 'varselect' firstly before 'normalize' step, an exception will be thrown to notice users they should do 'normalize' firstly.
