Quibbles with Activity 4.5

[Birone L]

From Q5 

Given what you know about how they operate, order these four algorithms accordng to the expected amount of overfitting, from greatest to least. Then confirm your intuition with a Weka experiment using the credit-g dataset.
 IBk < J48 < SMO ~ Logistic
 IBk < Logistic < SMO < J48
 Logistic < IBk < J48 < SMO
 J48 ~ Logistic < SMO

Shouldn't these be > signs? (eg if we're ranking change from accuracy in validation to fit of training set?)

(Also I was a little confused by what was meant by linearity in the earlier questions. We saw earlier that some methods produce blurry/probabilistic boundaries; is a blurry boundary  linear, even if it's constructed based on straight lines? And tree methods produce rules that seemed like the might be considered piecewise linear, since they're all straight, even if some are vertical & discontinuous in a mathematical sense. I could understand the rationales for the correct answers but couldn't get them right first time, despite relatively clear mental pictures of the kinds of boundaries the methods were producing. Perhaps it's just that my notion of linearity is hazy...)