Merge pull request #3 from TNO-Quantum/quandela

Add quandela solver

Author: RobertWezemanTNO

LICENSE

@@ -175,7 +175,7 @@
 
    END OF TERMS AND CONDITIONS
 
-   Copyright 2023 The Netherlands Organisation for Applied Scientific Research (TNO)
+   Copyright 2024 The Netherlands Organisation for Applied Scientific Research (TNO)
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.

README.md

@@ -4,17 +4,19 @@
 
 This repository contains python code to run the Q-score (Max-Cut and Max-Clique) benchmark on the following solver backends:
 
-- D-Wave devices and solvers, precisely its `Advantage_system4.1` and `DW_2000Q_6` QPU solvers, its `Simulated Annealing` and `qbsolv` classical solver, and its `hybrid` solver.
+- D-Wave QPU device, the `Advantage_system4.1` ~~and `DW_2000Q_6`~~ QPU system. (The `DW_2000Q_6` device no longer available.)
+- D-Wave classical solvers, its `Simulated Annealing` and `qbsolv` solver
+- D-Wave hybrid solver.
 - Gate-based hardware using QAOA on QuantumInspire and IBM hardware or simulators.
-- Simulator for Gaussian Boson Sampling, a form of photonic quantum computing.   
+- Gaussian Boson Sampling, a form of photonic quantum computing, both simulated an using the 12-mode Quandela QPU.   
 
 For an introduction to the Q-score, see the reference below.
-Q-score instances for Max-Cut or Max-Clique optimization problem can be run for different sizes and timeout limits. If no result is found within the allowed time limit, no objective result and a beta value of `0` is returned. Note that for the QPU solvers, the time limit considers embedding time only. The actual computation time will be slightly higher, but this difference will be in the order of milliseconds and will hence not influence the results. For similar reasons, for the photonic simulator, we only apply the time constraint to the classical runtime of the algorithm. To compute the Q-score, one runs for increasing graph size sufficiently many instances of the given code to check whether the average beta is larger than 0.2.
+Q-score instances for Max-Cut or Max-Clique optimization problem can be run for different sizes and timeout limits. If no result is found within the allowed time limit, no objective result and a beta value of `0` is returned. Note that for the QPU solvers, the time limit considers embedding time only. The actual computation time will be slightly higher, but this difference will be in the order of milliseconds and will hence not influence the results. For similar reasons, for the photonic simulator, we only apply the time constraint to the classical runtime of the algorithm. To compute the Q-score, one runs for increasing graph size sufficiently many instances of the given code to check whether the average beta is larger than `0.2`.
 
 This code was used to obtain results for the following papers:
 
 - ["Evaluating the Q-score of quantum annealers", Ward van der Schoot et al. (IEEE QSW 2022)](https://ieeexplore.ieee.org/document/9860191).
-- ["Q-score Max-Clique: The first metric evaluation on multiple computational paradigms", Ward van der Schoot et al. (preprint)](https://arxiv.org/abs/2302.00639)
+- ["Q-score Max-Clique: The first metric evaluation on multiple computational paradigms", Ward van der Schoot et al. (2023 arXiv)](https://arxiv.org/abs/2302.00639)
 
 ## Q-score introduction
 
@@ -63,7 +65,7 @@ Multiple Q-score instances for various sizes can be run as follows:
 
 To use this code we assume that the reader has installed the requirements and set up access to the required solvers. 
 
-Requirements can be installed using pip:
+Requirements can be installed using pip and have been tested for `python3.9` and `python3.10`:
 ```terminal
 python -m pip install -r requirements.txt
 ```
@@ -108,3 +110,13 @@ Example usage for the IBM Lima device:
 ```python
 python evaluate.py -p "max-clique" -s 5 -t 60 -n 1024 -solver "QAOA" -provider "ibm" -backend "ibmq_lima" 
 ```
+
+### Set up Quandela configuration
+
+To use the Quandela hardware backend, one need to create and Quandela Cloud account, which can be done [here](https://cloud.quandela.com/webide/login). 
+After creating your account you can store your API key inside the `configuration/_QUANDELA_API_KEY`:
+
+Example usage for the photonic Quandela Ascella hardware:
+```python
+python evaluate.py -p "max-clique" -s 3 -t 60 -n 1024 -solver "Photonic_quandela" -backend "qpu:ascella" 
+```

calculate_qscore.py

@@ -70,7 +70,8 @@ def calculate_qscore(
     for size in size_range:
         result, times = [], []
         exact_results = []
-        for _ in range(nb_instances_per_size):
+        for i in range(nb_instances_per_size):
+            print(f"START INSTANCE {i}")
             objective_result, _, time, G = main(
                 problem_type=problem_type,
                 size=size,

evaluate.py

@@ -13,6 +13,7 @@
 from run.run_dwave_qpu import run_dwave_qpu
 from run.run_hybrid import run_hybrid
 from run.run_photonic_simulated import run_photonic_simulated
+from run.run_photonic_quandela import run_photonic_quandela
 from run.run_QAOA import run_QAOA
 from run.run_qbsolv import run_qbsolv
 from run.run_SA import run_SA
@@ -69,7 +70,7 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument(
         "-provider",
         "--provider",
-        help="Name of hardware provider in case QAOA is selected.",
+        help="Name of hardware provider in case `QAOA` is selected.",
         choices=[
             "local simulator",
             "ibm",
@@ -81,7 +82,7 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument(
         "-backend",
         "--backend",
-        help="Name of backend in case QAOA is selected.",
+        help="Name of backend in case `QAOA` or `Photonic_quandela` is selected.",
         type=str,
         required=False,
     )
@@ -91,11 +92,11 @@ def parse_args() -> argparse.Namespace:
         help="String of the D-Wave solver.",
         choices=[
             "Advantage_system4.1",
-            "DW_2000Q_6",
             "hybrid",
             "qbsolv",
             "Simulated_Annealing",
             "Photonic_Simulation",
+            "Photonic_quandela",
             "QAOA",
         ],
         type=str,
@@ -127,7 +128,7 @@ def main(
         seed: random seed for reproducibility.
         num_reads: Number of reads/samples in case of a QPU or Simulated Annealing solver.
         provider: Name of hardware provider in case QAOA is selected.
-        backend: Name of backend in case QAOA is selected.
+        backend: Name of backend in case QAOA or photonic is selected.
 
     Returns:
         objective_result: solution to max-cut or max-clique.
@@ -142,16 +143,15 @@ def main(
     """
     if num_reads is None and solver in [
         "Advantage_system4.1",
-        "DW_2000Q_6",
         "Simulated_Annealing",
         "Photonic_Simulation",
+        "Photonic_quandela",
     ]:
         raise ValueError("num_reads has not been submitted while required by solver.")
 
     if seed is None:
         seed = np.random.randint(100000)
     G = nx.erdos_renyi_graph(size, 1 / 2, seed=seed)
-
     if solver == "QAOA":
         if problem_type == "max-cut":
             max_cut = Maxcut(G)
@@ -163,14 +163,20 @@ def main(
         start_time = time.time()
         objective_result = run_QAOA(qp, provider, backend)
         end_time = time.time()
-    elif solver == "Photonic_Simulation":
+    elif solver in ["Photonic_Simulation", "Photonic_quandela"]:
         if problem_type != "max-clique":
             raise ValueError(
-                "Photonic simulation solver can only be used for Max-Clique problem."
+                "Photonic solvers can only be used for Max-Clique problems."
+            )
+
+        if solver == "Photonic_Simulation":
+            objective_result, end_time, start_time = run_photonic_simulated(
+                G, size=size, n_samples=num_reads, timeout=timeout
+            )
+        elif solver == "Photonic_quandela":
+            objective_result, end_time, start_time = run_photonic_quandela(
+                G, size=size, backend=backend, n_samples=num_reads, timeout=timeout
             )
-        objective_result, end_time, start_time = run_photonic_simulated(
-            G, size=size, n_samples=num_reads, timeout=timeout
-        )
     else:
         # Create qubo:
         if problem_type == "max-cut":
@@ -183,7 +189,7 @@ def main(
             )
 
         # Solve problem instance
-        if solver in ["Advantage_system4.1", "DW_2000Q_6"]:
+        if solver == "Advantage_system4.1":
             start_time = time.time()
             objective_result = run_dwave_qpu(Q, size, solver, num_reads, timeout)
             end_time = time.time()

requirements.txt

@@ -1,9 +1,31 @@
+# generic dependencies
 numpy==1.22.1
 networkx==2.8.8
+matplotlib==3.5.2
+
+# d-wave dependencies
+dwave-cloud-client==0.9.3
+dwave-greedy==0.2.2
+dwave-hybrid==0.6.5
+dwave-inspector==0.2.9
+dwave-neal==0.5.9
+dwave-networkx==0.8.11
 dwave-ocean-sdk==4.4.0
+dwave-preprocessing==0.3.2
+dwave-qbsolv==0.3.4
+dwave-system==1.12.0
+dwave-tabu==0.4.3
+dwavebinarycsp==0.1.4
+pydantic==1.10.4
+
+# photonic dependencies
 strawberryfields==0.23.0
-matplotlib==3.5.2
-qiskit[optimization]==0.39.4
-quantuminspire==2.0.0
+perceval-quandela==0.8.1
 python-dotenv==0.21.0
+
+# qaoa dependencies
+qiskit[optimization]==0.39.4
+quantuminspire==2.2.1
+
+# testing dependencies
 pytest==7.2.1