[Feature] Add sample, Use single precision by default

[dominikandreasseitz]

Closes #18

[RolandMacDoland]

Hey @dominikandreasseitz, I'm working on fixing this PR. One thing that came up is that in test_hamevo_general while constructing the Hamiltonian:

(Pdb) hamevo
_HamiltonianEvolution(generator_name='I', target=((0, 1),), control=((None,),))

Just wondering what is the rationale behind the setting of target and control. What I understand is that:

    @staticmethod
    def parse_idx(
        idx: Tuple,
    ) -> Tuple:
        if isinstance(idx, (int, np.int64)):
            return ((idx,),)
        elif isinstance(idx, tuple):
            return (idx,)
        else:
            return (idx.astype(int),)

which makes target and control being tuples of tuples. Is this what is really intended ? For instance, defaults for control are (None,) which seems contradictory with the setting in parse_idx. Any guidance on this ? Thanks a lot :)

[dominikandreasseitz]

Hi @Roland-djee , the reason for the tuple of tuples design choice lies in the apply_gate function which can be either applied to a single gate or an arbitrary number of gates. To do so, the operators and supports are collected in a single tuple and then get applied in a single lambda expression. The "collection" part assumes a tuple of tuples for supports so thats the story behind that. I get its a bit counterintuitive but its part of legacy horqrux so i was hesitant to change it since the function is quite useful

[chMoussa]

Hey @RolandMacDoland I think I managed to fix this MR. What do you think?

[RolandMacDoland]

Hey @RolandMacDoland I think I managed to fix this MR. What do you think?

Oh, wow cool thanks. Please feel free to request a review :)

[RolandMacDoland]

Thanks @chMoussa. Looking good. Few comments for the sake of discussion.

[RolandMacDoland]

LGTM. Minor renaming quibble. I think it is conventional in torch to have vparams and fparams to disambiguate.