Two methods that are acceptable are:

1) if class zero is more common than class 1, predict a probability of 0 for all points and
calculate the RMSE of this prediction on the data set.  (if class 1 is more common predict prob 1
for all points).

2) if the number of class 0 is a and the number of class 1 is b in the data, predict
probability a/(a+b) for each point and calculate RMSE of this prediction on the data set.