You are right when all the odd values are observed. But in this case we want likelihood than an observation has odd value. So likelihood function is { [P(lambda=odd) ]^1}*{ [ P(lambda= not odd) ]^0 }.
