Extensible sieve of Eratosthenes

I was creating a Carmichael triplets generator and needed a function to tell me if a number was prime. It wasn't immediately clear what the largest prime for checking would be so I decided to look into creating an extensible prime generator - If given a number to check for prime-ness that was outside its range, it would attempt to extend its range.

I stopped modifying the class as soon as it was working as the Carmichael generator task then ran in acceptable time.

class Isprime():
    '''
    Extensible sieve of Erastosthenes
    
    >>> isprime.check(11)
    True
    >>> isprime.multiples
    {2, 4, 6, 8, 9, 10}
    >>> isprime.primes
    [2, 3, 5, 7, 11]
    >>> isprime(13)
    True
    >>> isprime.multiples
    {2, 4, 6, 7, 8, 9, 10, 11, 12, 14, 15, 16, 18, 20, 21, 22}
    >>> isprime.primes
    [2, 3, 5, 7, 11, 13, 17, 19]
    >>> isprime.nmax
    22
    >>> 
    '''
    multiples = {2}
    primes = [2]
    nmax = 2

    def __init__(self, nmax):
        if nmax > self.nmax:
            self.check(nmax)

    def check(self, n):
        if type(n) == float:
            if not n.is_integer(): return False
            n = int(n)
        multiples = self.multiples
        if n <= self.nmax:
            return n not in multiples
        else:
            # Extend the sieve
            primes, nmax = self.primes, self.nmax
            newmax = max(nmax*2, n)
            for p in primes:
                multiples.update(range(p*((nmax + p + 1) // p), newmax+1, p))
            for i in range(nmax+1, newmax+1):
                if i not in multiples:
                    primes.append(i)
                    multiples.update(range(i*2, newmax+1, i))
            self.nmax = newmax
            return n not in multiples

    __call__ = check

Now I wouldn't be surprised if some of the loop indices are not optimum and are maybe recomputing a few values that don't need to be, and this code is only efficient enough for my purposes; (a couple of Rosetta Code entries in other languages state that they are based on the Python but make changes so that is an indication).