>>> np.issubdtype(np.float64,np.float32) #this is ok
False
>>> np.issubdtype('float64','float32') #but this is not ok
True

According to the docs, the arguments of issubdtype can be "dtype or string representing a typecode", but it fails in the second case.