I noticed that if I have a long table (relation) rel, then calling len(rel) can be quite slow. It takes about 2 seconds on our setup to return if the length is say, 300000 tuples. Here's the code in RelationalOperand:

    def __len__(self):
        """
        number of tuples in the relation.
        """
        return U().aggr(self, n='count(*)').fetch1('n')

After looking at the code for RelationalOperand.preview(), I noticed that it uses len(self.proj()) instead of len(self), which returns the same result in about half the time. Could __len__ be sped up? Maybe just replace it with len(self.proj())?