In Bayes' theorem, is it necessary that the Likelihood function, say $P(X|\theta)$ must be of a discrete probability mass function? I ask because it seems like this is saying that what is the probability of observing $X$ given $\theta$, but if this was a continuous PDF, then the probability of observing exactly $X$ would always be 0. All of the examples that I've seen in my textbook always give the this function has that of a binomial distribution or some other discrete PMF, but it seems to restrictive to me.
Thanks.
Indeed it depends on the distribution model. The likelihood function for a absolutely continuous random variable's distribution parameter corresponds to a conditioned probability densitity function.
$$\mathcal L(\theta\mid x)=f(x\mid \theta)$$