GLM model coefficients that are unused (due to lambda or whatever)...model returns ''". Not like h2o1, which didn't return anything.
On h2o1, if a column was ignored or the GLM resulted in that coefficient being not used (say due to regularization forcing it to zero)...GLM wouldn't return that coefficient's name or value in the json response
in h2o-dev, we're getting all the coefficients in the 2 dim table now.
if the coefficient is not used, the returned value is "" ...i..e. empty quoted string.
If that's the intent, then we need to define that as a legal response. I'm failing right now because I'm expecting a number (quoted for now) or quoted Nan or Infinity. Not empty string.
talked to tomas about this. Not sure what the plan is.
Also, there's an old coefficients_magnitude field in the GLM model build response that is no longer used/valid with the two dim table. Should remove that
maybe it's inconsistent.
I just noticed here in covtype I get 0's for some coefficients (I think h2o1 used to drop those)
I used to argue that having the 0's was better
C1: 2.08246e-02 C2: -3.34284e-04 C3: -1.65310e-02 C4: -5.34450e-04 C5: 5.71817e-03 C6: -1.59905e-04 C7: -3.84853e-03 C8: -1.86324e-02 C9: -2.59158e-03 C10: 2.29180e-04 C11: 1.65118e-02 C12: 7.76315e-04 C13: -1.73881e-02 C14: 0.00000e+00 C15: 0.00000e+00 C16: 0.00000e+00 C17: 0.00000e+00 C18: 0.00000e+00 C19: 0.00000e+00 C20: 0.00000e+00 C21: 0.00000e+00 C22: 0.00000e+00 C23: 0.00000e+00 C24: 0.00000e+00 C25: 0.00000e+00 C26: 0.00000e+00 C27: 0.00000e+00 C28: 0.00000e+00 C29: 0.00000e+00 C30: 0.00000e+00 C31: 0.00000e+00 C32: 0.00000e+00 C33: 0.00000e+00 C34: 0.00000e+00 C35: 0.00000e+00 C36: -2.72968e-03 C37: -3.07880e-03 C38: -1.43480e-03 C39: 0.00000e+00 C40: 0.00000e+00 C41: -5.34080e-04 C42: 0.00000e+00 C43: -2.91639e-03 C44: -4.28268e-04 C45: -3.45066e-03 C46: -1.18132e-02 C47: -7.63632e-03 C48: 0.00000e+00 C49: -5.85888e-03 C50: 0.00000e+00 C51: -1.22549e-03 C52: -1.88317e-02 C53: -2.16577e-02 C55: -5.37986e-03 Intercept: -6.68887e+01