Demos

[cgogolin]

Hi Maria, As a warm-up for implementing unit-test I have played around with the demos and have some qusestions/feedback:

• 1-qubit_rotation.py: Works beautifully. The convergence is very fast, or at least the number format for printing the cost is such that one cannot see the convergence. Maybe it would be good to switch to a scientific number format?
• 2a-variational_classifier-iris.py: This uses QubitStateVector and therefore doesn't work with ProjectQ. Can we modify the demo to not use QubitStateVector?
• 2b-variational_classifier-parity.py: Works but is rather slow. Is that expected? I have turned some of the comments into print statements (maybe that is a good idea for the other demos as well?) so that the user sees some output before the first iteration has completed.
• 3a-variational_quantum_eigensolver-quantumopt.py: Works. I have increased the number of iterations slightly, so that the cost actually converges to -0.1.
• 3b-variational_quantum_eigensolver-classopt.py: Works, but here the cost seems to never converge. I think this is because the cost function is not bounded. It can be made arbitrarily small by simply multiplying the weights with a constant. This is maybe not so nice? Maybe we can force the sum of the squared weights to be at most 1? Also the docstring in the beginning suggests that we optimize a state for a fixed Hamiltonian. Looking at the code I get the impression that the opposite is happening: We are optimizing a Hamiltionian for a fixed state.

Note: Please be aware that due to the renaming and refactoring you will likely have to uninstall all old versions of openqml and openqml_pq and then install first the latest master of openqml and then the latest master of openqml_pq to make everything work.

[mariaschuld]

Hey Christian. Thanks for the comments! Let me comment back. PS: I think you did not look at the examples branch, did you? Because some issues are solved there and the examples majorly polished...

For 1 - This is the example where the gradient descent optimizer gets stuck in a saddle point. Should converge to -1, not 0. Let me open a separate issue for this. I decided against scientific notation because the eye traces convergence better with fixed point notation.

2a (now 3a) - Solved in principle. I decomposed stateprep but there is a bug somewhere. On my to do list. 2b (now 3b) - State prep is an issue. I need BasisState(...) to work for this. In short, since in qnode I don't have access to unwrapped kwargs, such as the binary strings I input, I cannot do the "if x[i] == 1 flip the ith qubit" command. Will open another issue. 3a (now 2a) - Yes, should work. Need a nicer initial state though to make opt landscape more interesting. 1/sqrt(3)(|00> + |10> + |11>) worked well, but need to figure out how to prepare this. Just polishing jupyter notebook. 3b (now 2b) - Yes, cost is unbounded. Tried square, but then the quantum example stops being so nice. Not sure yet.

-> Note: I tried to keep track of developments by pip installing openqml-pq and openqml every morning...

[mariaschuld]

Let me open issues for the things I'd need your help with :)

[cgogolin]

Indeed, I made a mistake here. Let me close this...