Allow fixed parameters in GeneralEncoder

[nikhilkhatri]

This PR adds the ability to fix parameters of some parameterised gates in a GeneralEncoder.

This allows, for example, a single parameterised gate to have a fixed parameter, shared across all items in the batch, while all other parameters can be dynamically provided. The change is fully backwards compatible, and no changes need to be made to existing encoder definitions.

This supports ansatze which have some fixed gates which aren't Paulis or other named operators.
I expect this could already be achieved by subclassing existing operators, but I think this imposes less effort on the user.

Opening this to gauge interest. Shall update documentation and tests pending feedback.

>>> import torchquantum as tq
>>> import torch
>>> 
>>> 
>>> enc1 = [{'input_idx': [0], 'func': 'rx', 'wires': [0]},
...        {'input_idx': [1], 'func': 'rx', 'wires': [1]},
...        {'input_idx': [2], 'func': 'rx', 'wires': [2]},
...        {'input_idx': [3], 'func': 'rx', 'wires': [3]}]
>>> 
>>> # Same as above, but now with 0.3 provided to the first RX gate always
>>> enc2 = [{'params': [0.3], 'func': 'rx', 'wires': [0]},
...        {'input_idx': [0], 'func': 'rx', 'wires': [1]},
...        {'input_idx': [1], 'func': 'rx', 'wires': [2]},
...        {'input_idx': [2], 'func': 'rx', 'wires': [3]}]
>>> 
>>> 
>>> e1 = tq.GeneralEncoder(enc1)
>>> e2 = tq.GeneralEncoder(enc2)
>>> 
>>> qdev = tq.QuantumDevice(4)
>>> 
>>> 
>>> params1 = torch.tensor([[0.3, 0.4, 0.5, 0.6], [0.3, 0, 0, 0]])
>>> params2 = torch.tensor([[0.4, 0.5, 0.6], [0, 0, 0]])
>>> 
>>> qdev.reset_states(2)
>>> e1(qdev, params1)
>>> print(qdev.get_states_1d())
tensor([[ 0.8970+0.0000j,  0.0000-0.2775j,  0.0000-0.2290j, -0.0709+0.0000j,
          0.0000-0.1818j, -0.0562+0.0000j, -0.0464+0.0000j,  0.0000+0.0144j,
          0.0000-0.1356j, -0.0419+0.0000j, -0.0346+0.0000j,  0.0000+0.0107j,
         -0.0275+0.0000j,  0.0000+0.0085j,  0.0000+0.0070j,  0.0022+0.0000j],
        [ 0.9888+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,
          0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,
          0.0000-0.1494j,  0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,
          0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j]])
>>> 
>>> qdev.reset_states(2)
>>> e2(qdev, params2)
>>> print(qdev.get_states_1d())
tensor([[ 0.8970+0.0000j,  0.0000-0.2775j,  0.0000-0.2290j, -0.0709+0.0000j,
          0.0000-0.1818j, -0.0562+0.0000j, -0.0464+0.0000j,  0.0000+0.0144j,
          0.0000-0.1356j, -0.0419+0.0000j, -0.0346+0.0000j,  0.0000+0.0107j,
         -0.0275+0.0000j,  0.0000+0.0085j,  0.0000+0.0070j,  0.0022+0.0000j],
        [ 0.9888+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,
          0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,
          0.0000-0.1494j,  0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,
          0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j,  0.0000+0.0000j]])

[nikhilkhatri]

Hi @GenericP3rson, could I get a first-pass review on this please?

[01110011011101010110010001101111]

Yup, on an initial review, this looks solid to me! I want to run it through the rest of the team before officially merging; we’ll meet sometime this week so you’ll get a more definite response by then!

Also, as you mentioned, if you could add some tests, that would be great!

[nikhilkhatri]

I've added the tests now @GenericP3rson !