About the Features of Model

[xiandong79]

In Ernest paper, it seems that it uses [machine, scale] to generate [feature array].
In experiment design, it seems that it uses [machine, scale] to generate [feature array] and selects the 'best information' training points.
In prediction step, however, it uses [cores, scale] to generate the [feature array] and fits the final time/machine model.

Sorry to bother you guys. @shivaram