How to overload __init__ method based on argument type?

Excellent question. I've tackled this problem as well, and while I agree that "factories" (class-method constructors) are a good method, I would like to suggest another, which I've also found very useful:

Here's a sample (this is a read method and not a constructor, but the idea is the same):

def read(self, str=None, filename=None, addr=0):
    """ Read binary data and return a store object. The data
        store is also saved in the interal 'data' attribute.

        The data can either be taken from a string (str 
        argument) or a file (provide a filename, which will 
        be read in binary mode). If both are provided, the str 
        will be used. If neither is provided, an ArgumentError 
        is raised.
    """
    if str is None:
        if filename is None:
            raise ArgumentError('Please supply a string or a filename')

        file = open(filename, 'rb')
        str = file.read()
        file.close()
    ...
    ... # rest of code

The key idea is here is using Python's excellent support for named arguments to implement this. Now, if I want to read the data from a file, I say:

obj.read(filename="blob.txt")

And to read it from a string, I say:

obj.read(str="\x34\x55")

This way the user has just a single method to call. Handling it inside, as you saw, is not overly complex

A much neater way to get 'alternate constructors' is to use classmethods. For instance:

>>> class MyData:
...     def __init__(self, data):
...         "Initialize MyData from a sequence"
...         self.data = data
...     
...     @classmethod
...     def fromfilename(cls, filename):
...         "Initialize MyData from a file"
...         data = open(filename).readlines()
...         return cls(data)
...     
...     @classmethod
...     def fromdict(cls, datadict):
...         "Initialize MyData from a dict's items"
...         return cls(datadict.items())
... 
>>> MyData([1, 2, 3]).data
[1, 2, 3]
>>> MyData.fromfilename("/tmp/foobar").data
['foo\n', 'bar\n', 'baz\n']
>>> MyData.fromdict({"spam": "ham"}).data
[('spam', 'ham')]

The reason it's neater is that there is no doubt about what type is expected, and you aren't forced to guess at what the caller intended for you to do with the datatype it gave you. The problem with isinstance(x, basestring) is that there is no way for the caller to tell you, for instance, that even though the type is not a basestring, you should treat it as a string (and not another sequence.) And perhaps the caller would like to use the same type for different purposes, sometimes as a single item, and sometimes as a sequence of items. Being explicit takes all doubt away and leads to more robust and clearer code.