This is a question more for advice and a suggested starting point than anything else (though anything else is cool as well )
The data that I have is something like this - 1,000,000 data points of which about 9/10 are 0's, the rest are strictly positive.
The way that is best to understand this process is as a two step process: 1) Something like a binomial distribution says if a event occurs or not. 2) If the event occurs then we take a random draw from the continuous distribution over a positive range (something like an extreme value distribution).
I am curious about the best way of bringing these two separate distributions into a single distribution; so if any one has any suggestions they will be more than welcome!
