It looks like the initial intercept term is 1 only in the addIntercept && numOfLinearPredictor == 1 case. It does seem inconsistent; since it's just an initial weight it may not matter to the final converged value. You can see a few notes in the class about how numOfLinearPredictor == 1 is handled a bit inconsistently and how a smarter choice of initial intercept could help convergence. So I don't know if this rises to the level of bug but I don't know that the difference is on purpose.
On Thu, Feb 5, 2015 at 5:40 PM, jamborta <jambo...@gmail.com> wrote: > hi all, > > I have been going through the GeneralizedLinearAlgorithm to understand how > intercepts are handled in regression. Just noticed that the initial setting > for the intercept is set to one (whereas the initial setting for the rest of > the coefficients is set to zero) using the same piece of code that adds the > 1 in front of each line in the data. Is this a bug? > > thanks, > > > > -- > View this message in context: > http://apache-spark-user-list.1001560.n3.nabble.com/one-is-the-default-value-for-intercepts-in-GeneralizedLinearAlgorithm-tp21525.html > Sent from the Apache Spark User List mailing list archive at Nabble.com. > > --------------------------------------------------------------------- > To unsubscribe, e-mail: user-unsubscr...@spark.apache.org > For additional commands, e-mail: user-h...@spark.apache.org > --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@spark.apache.org For additional commands, e-mail: user-h...@spark.apache.org
