Merge pull request #1202 from qiboteam/pytorch_backend

`pytorch` backend

Author: renatomello

doc/source/api-reference/qibo.rst

@@ -2375,7 +2375,7 @@ The user can switch backends using
     qibo.set_backend("numpy")
 
 before creating any circuits or gates. The default backend is the first available
-from ``qibojit``, ``tensorflow``, ``numpy``.
+from ``qibojit``, ``pytorch``, ``tensorflow``, ``numpy``.
 
 Some backends support different platforms. For example, the qibojit backend
 provides two platforms (``cupy`` and ``cuquantum``) when used on GPU.
@@ -2456,4 +2456,8 @@ Alternatively, a Clifford circuit can also be executed starting from the :class:
 Cloud Backends
 ^^^^^^^^^^^^^^
 
-Additional backends, that support the remote execution of quantum circuits through cloud service providers, such as IBM and QRC-TII, are provided by the optional qibo plugin `qibo-cloud-backends <https://github.com/qiboteam/qibo-cloud-backends>`_. For more information please refer to the `official documentation <https://qibo.science/qibo-cloud-backends/stable/>`_.
+Additional backends that support the remote execution of quantum circuits through
+cloud service providers, such as IBM and QRC-TII, are provided by the optional qibo plugin
+`qibo-cloud-backends <https://github.com/qiboteam/qibo-cloud-backends>`_.
+For more information please refer to the
+`official documentation <https://qibo.science/qibo-cloud-backends/stable/>`_.

doc/source/getting-started/backends.rst

@@ -36,10 +36,10 @@ if the corresponding packages are installed, following the hierarchy below:
 
 * :ref:`installing-numpy`: a lightweight quantum simulator shipped with the :ref:`installing-qibo` base package. Use this simulator if your CPU architecture is not supported by the other backends. Please note that the simulation performance is quite poor in comparison to other backends.
 * :ref:`installing-qibojit`: an efficient simulation backend for CPU, GPU and multi-GPU based on just-in-time (JIT) compiled custom operators. Install this package if you need to simulate quantum circuits with large number of qubits or complex quantum algorithms which may benefit from computing parallelism.
-* `qibotn <https://qibo.science/qibotn/stable/>`_: an interface to Tensor Networks simulation algorithms designed for GPUs and multi-node CPUs. This backend makes possible scaling quantum circuit simulation to a larger number of qubits.
 * :ref:`installing-tensorflow`: a pure TensorFlow implementation for quantum simulation which provides access to gradient descent optimization and the possibility to implement classical and quantum architectures together. This backend is not optimized for memory and speed, use :ref:`installing-qibojit` instead.
 * :ref:`installing-pytorch`: a pure PyTorch implementation for quantum simulation which provides access to gradient descent optimization and the possibility to implement classical and quantum architectures together. This backend is not optimized for memory and speed, use :ref:`installing-qibojit` instead.
 * :ref:`clifford <Clifford>`: a specialized backend for the simulation of quantum circuits with Clifford gates. This backend uses :ref:`installing-qibojit` and/or :ref:`installing-numpy`.
+* `qibotn <https://qibo.science/qibotn/stable/>`_: an interface to Tensor Networks simulation algorithms designed for GPUs and multi-node CPUs. This backend makes possible scaling quantum circuit simulation to a larger number of qubits.
 
 The default backend that is used is the first available from the above list.
 The user can switch to a different using the ``qibo.set_backend`` method

examples/adiabatic/trotter_error.py

@@ -61,9 +61,9 @@ def main(nqubits, hfield, T, save):
     ]
     alphas = [1.0, 0.7, 0.4]
     labels = [
-        r"$\delta t ^{}$".format(exponent - 1),
-        r"$\delta t ^{}$".format(exponent),
-        r"$\delta t ^{}$".format(exponent + 1),
+        f"$\\delta t ^{exponent - 1}$",
+        f"$\\delta t ^{exponent}$",
+        f"$\\delta t ^{exponent - 1}$",
     ]
 
     plt.figure(figsize=(7, 4))

examples/adiabatic3sat/functions.py

@@ -17,7 +17,7 @@ def read_file(file_name, instance):
         solution (list): list of the correct outputs of the instance for testing.
         clauses (list): list of all clauses, with the qubits each clause acts upon.
     """
-    file = open("../data3sat/{q}bit/n{q}i{i}.txt".format(q=file_name, i=instance))
+    file = open(f"../data3sat/{file_name}bit/n{file_name}i{instance}.txt")
     control = list(map(int, file.readline().split()))
     solution = list(map(str, file.readline().split()))
     clauses = [list(map(int, file.readline().split())) for _ in range(control[1])]
@@ -100,12 +100,12 @@ def plot(qubits, ground, first, gap, dt, T):
     plt.title("Energy during adiabatic evolution")
     ax.legend()
     fig.tight_layout()
-    fig.savefig("{}_qubits_energy.png".format(qubits), dpi=300, bbox_inches="tight")
+    fig.savefig(f"{qubits}_qubits_energy.png", dpi=300, bbox_inches="tight")
     fig, ax = plt.subplots()
     ax.plot(times, gap, label="gap energy", color="C0")
     plt.ylabel("energy")
     plt.xlabel("schedule")
     plt.title("Energy during adiabatic evolution")
     ax.legend()
     fig.tight_layout()
-    fig.savefig("{}_qubits_gap.png".format(qubits), dpi=300, bbox_inches="tight")
+    fig.savefig(f"{qubits}_qubits_gap.png", dpi=300, bbox_inches="tight")

examples/adiabatic3sat/main.py

@@ -48,9 +48,8 @@ def main(nqubits, instance, T, dt, solver, plot, dense, params, method, maxiter)
     print("-" * 20 + "\n")
     if plot and nqubits >= 14:
         print(
-            "Currently not possible to calculate gap energy for {} qubits."
-            "\n Proceeding to adiabatic evolution without plotting data.\n"
-            "".format(nqubits)
+            f"Currently not possible to calculate gap energy for {nqubits} qubits."
+            + "\n Proceeding to adiabatic evolution without plotting data.\n"
         )
         plot = False
     if plot and method is not None:
@@ -97,9 +96,9 @@ def main(nqubits, instance, T, dt, solver, plot, dense, params, method, maxiter)
     output_dec = (np.abs(final_state) ** 2).argmax()
     max_output = "{0:0{bits}b}".format(output_dec, bits=nqubits)
     max_prob = (np.abs(final_state) ** 2).max()
-    print("Exact cover instance with {} qubits.\n".format(nqubits))
+    print(f"Exact cover instance with {nqubits} qubits.\n")
     if solution:
-        print("Known solution: {}\n".format("".join(solution)))
+        print(f"Known solution: {''.join(solution)}\n")
     print("-" * 20 + "\n")
     print(
         f"Adiabatic evolution with total time {T}, evolution step {dt} and "

examples/benchmarks/circuits.py

@@ -67,7 +67,7 @@ def CircuitFactory(nqubits, circuit_name, accelerators=None, **kwargs):
         circuit = models.QFT(nqubits, accelerators=accelerators)
     else:
         if circuit_name not in _CIRCUITS:
-            raise KeyError("Unknown benchmark circuit type {}." "".format(circuit_name))
+            raise KeyError(f"Unknown benchmark circuit type {circuit_name}.")
         circuit = models.Circuit(nqubits, accelerators=accelerators)
         circuit.add(_CIRCUITS.get(circuit_name)(nqubits, **kwargs))
     return circuit

examples/benchmarks/main.py

@@ -53,14 +53,14 @@ def limit_gpu_memory(memory_limit=None):
         print("\nNo GPU memory limiter used.\n")
         return
 
-    print("\nAttempting to limit GPU memory to {}.\n".format(memory_limit))
+    print(f"\nAttempting to limit GPU memory to {memory_limit}.\n")
     gpus = tf.config.list_physical_devices("GPU")
     for gpu in tf.config.list_physical_devices("GPU"):
         config = tf.config.experimental.VirtualDeviceConfiguration(
             memory_limit=memory_limit
         )
         tf.config.experimental.set_virtual_device_configuration(gpu, [config])
-        print("Limiting memory of {} to {}.".format(gpu.name, memory_limit))
+        print(f"Limiting memory of {gpu.name} to {memory_limit}.")
     print()
 
 

examples/benchmarks/utils.py

@@ -6,12 +6,12 @@ class BenchmarkLogger(list):
     def __init__(self, filename=None):
         self.filename = filename
         if filename is not None and os.path.isfile(filename):
-            print("Extending existing logs from {}.".format(filename))
+            print(f"Extending existing logs from {filename}.")
             with open(filename) as file:
                 super().__init__(json.load(file))
         else:
             if filename is not None:
-                print("Creating new logs in {}.".format(filename))
+                print(f"Creating new logs in {filename}.")
             super().__init__()
 
     def dump(self):
@@ -20,7 +20,7 @@ def dump(self):
                 json.dump(list(self), file)
 
     def __str__(self):
-        return "\n".join("{}: {}".format(k, v) for k, v in self[-1].items())
+        return "\n".join(f"{k}: {v}" for k, v in self[-1].items())
 
 
 def parse_accelerators(accelerators):

examples/grover3sat/functions.py

@@ -15,7 +15,7 @@ def read_file(file_name, instance):
         solution (list): list of the correct outputs of the instance for testing.
         clauses (list): list of all clauses, with the qubits each clause acts upon.
     """
-    file = open("../data3sat/{q}bit/n{q}i{i}.txt".format(q=file_name, i=instance))
+    file = open(f"../data3sat/{file_name}bit/n{file_name}i{instance}.txt")
     control = list(map(int, file.readline().split()))
     solution = list(map(str, file.readline().split()))
     clauses = [list(map(int, file.readline().split())) for _ in range(control[1])]

examples/grover3sat/main.py

@@ -18,16 +18,16 @@ def main(nqubits, instance):
     qubits = control[0]
     clauses_num = control[1]
     steps = int((np.pi / 4) * np.sqrt(2**qubits))
-    print("Qubits encoding the solution: {}\n".format(qubits))
-    print("Total number of qubits used:  {}\n".format(qubits + clauses_num + 1))
+    print(f"Qubits encoding the solution: {qubits}\n")
+    print(f"Total number of qubits used: {qubits + clauses_num + 1}\n")
     q, c, ancilla, circuit = functions.create_qc(qubits, clauses_num)
     circuit = functions.grover(circuit, q, c, ancilla, clauses, steps)
     result = circuit(nshots=100)
     frequencies = result.frequencies(binary=True, registers=False)
     most_common_bitstring = frequencies.most_common(1)[0][0]
-    print("Most common bitstring: {}\n".format(most_common_bitstring))
+    print(f"Most common bitstring: {most_common_bitstring}\n")
     if solution:
-        print("Exact cover solution:  {}\n".format("".join(solution)))
+        print(f"Exact cover solution: {''.join(solution)}\n")
 
 
 if __name__ == "__main__":

examples/hash-grover/main.py

@@ -23,11 +23,9 @@ def main(h_value, collisions, b):
     h = "{0:0{bits}b}".format(h_value, bits=b)
     if len(h) > 8:
         raise ValueError(
-            "Hash should be at maximum an 8-bit number but given value contains {} bits.".format(
-                len(h)
-            )
+            f"Hash should be at maximum an 8-bit number but given value contains {len(h)} bits."
         )
-    print("Target hash: {}\n".format(h))
+    print(f"Target hash: {h}\n")
     if collisions:
         grover_it = int(np.pi * np.sqrt((2**8) / collisions) / 4)
         result = functions.grover(q, constant_1, constant_2, rot, h, grover_it)
@@ -36,19 +34,19 @@ def main(h_value, collisions, b):
         print("Preimages:")
         for i in most_common:
             if functions.check_hash(q, i[0], h, constant_1, constant_2, rot):
-                print("   - {}\n".format(i[0]))
+                print(f"   - {i[0]}\n")
             else:
                 print(
                     "   Incorrect preimage found, number of given collisions might not match.\n"
                 )
-        print("Total iterations taken: {}\n".format(grover_it))
+        print(f"Total iterations taken: {grover_it}\n")
     else:
         measured, total_iterations = functions.grover_unknown_M(
             q, constant_1, constant_2, rot, h
         )
         print("Solution found in an iterative process.\n")
-        print("Preimage: {}\n".format(measured))
-        print("Total iterations taken: {}\n".format(total_iterations))
+        print(f"Preimage: {measured}\n")
+        print(f"Total iterations taken: {total_iterations}\n")
 
 
 if __name__ == "__main__":

examples/unary/functions.py

@@ -482,7 +482,7 @@ def paint_prob_distribution(bins, prob_sim, S0, sig, r, T):
     ax.plot(x, y, label="PDF", color="black")
     plt.ylabel("Probability")
     plt.xlabel("Option price")
-    plt.title("Option price distribution for {} qubits ".format(bins))
+    plt.title(f"Option price distribution for {bins} qubits ")
     ax.legend()
     fig.tight_layout()
     fig.savefig("Probability_distribution.png")

examples/unary/main.py

@@ -15,7 +15,7 @@ def main(data, bins, M, shots):
 
     # Generate the probability distribution plots
     fun.paint_prob_distribution(bins, prob_sim, S0, sig, r, T)
-    print("Histogram printed for unary simulation with {} qubits.\n".format(bins))
+    print(f"Histogram printed for unary simulation with {bins} qubits.\n")
 
     # Create circuit to compute the expected payoff
     circuit, S = fun.load_payoff_quantum_sim(bins, S0, sig, r, T, K)
@@ -28,14 +28,14 @@ def main(data, bins, M, shots):
 
     # Finding differences between exact value and quantum approximation
     error = fun.diff_qu_cl(qu_payoff_sim, cl_payoff)
-    print("Exact value of the expected payoff:      {}\n".format(cl_payoff))
-    print("Expected payoff from quantum simulation: {}\n".format(qu_payoff_sim))
-    print("Percentage error: {} %\n".format(error))
+    print(f"Exact value of the expected payoff:      {cl_payoff}\n")
+    print(f"Expected payoff from quantum simulation: {qu_payoff_sim}\n")
+    print(f"Percentage error: {error} %\n")
     print("-" * 60 + "\n")
 
     # Applying amplitude estimation
     a_s, error_s = fun.amplitude_estimation(bins, M, data)
-    print("Amplitude estimation with a total of {} runs.\n".format(M))
+    print(f"Amplitude estimation with a total of {M} runs.\n")
     fun.paint_AE(a_s, error_s, bins, M, data)
     print("Amplitude estimation result plots generated.")
 

examples/variational_classifier/main.py

@@ -72,7 +72,7 @@ def main(nclasses, nqubits, nlayers, nshots, training, RxRzRx, method):
                 path_angles = (
                     LOCAL_FOLDER
                     / "data"
-                    / "optimal_angles_ry_{}q_{}l.npy".format(nqubits, nlayers)
+                    / f"optimal_angles_ry_{nqubits}q_{nlayers}l.npy"
                 )
                 optimal_angles = np.load(path_angles)
             except:
@@ -84,7 +84,7 @@ def main(nclasses, nqubits, nlayers, nshots, training, RxRzRx, method):
                 path_angles = (
                     LOCAL_FOLDER
                     / "data"
-                    / "optimal_angles_rxrzrx_{}q_{}l.npy".format(nqubits, nlayers)
+                    / f"optimal_angles_rxrzrx_{nqubits}q_{nlayers}l.npy"
                 )
                 optimal_angles = np.load(path_angles)
             except:
@@ -111,9 +111,7 @@ def main(nclasses, nqubits, nlayers, nshots, training, RxRzRx, method):
                 method=method,
             )
             path_angles = (
-                LOCAL_FOLDER
-                / "data"
-                / "optimal_angles_ry_{}q_{}l.npy".format(nqubits, nlayers)
+                LOCAL_FOLDER / "data" / f"optimal_angles_ry_{nqubits}q_{nlayers}l.npy"
             )
             np.save(
                 path_angles,
@@ -138,7 +136,7 @@ def main(nclasses, nqubits, nlayers, nshots, training, RxRzRx, method):
             path_angles = (
                 LOCAL_FOLDER
                 / "data"
-                / "optimal_angles_rxrzrx_{}q_{}l.npy".format(nqubits, nlayers)
+                / f"optimal_angles_rxrzrx_{nqubits}q_{nlayers}l.npy"
             )
             np.save(
                 path_angles,
@@ -170,14 +168,12 @@ def main(nclasses, nqubits, nlayers, nshots, training, RxRzRx, method):
     ]
 
     print(
-        "Train set | # Clases: {} | # Qubits: {} | # Layers: {} | Accuracy: {}".format(
-            nclasses, nqubits, nlayers, qc.Accuracy(labels_train, predictions_train)
-        )
+        f"Train set | # Clases: {nclasses} | # Qubits: {nqubits} | # Layers: {nlayers} | "
+        + f"Accuracy: {qc.Accuracy(labels_train, predictions_train)}"
     )
     print(
-        "Test set  | # Clases: {} | # Qubits: {} | # Layers: {} | Accuracy: {}".format(
-            nclasses, nqubits, nlayers, qc.Accuracy(labels_test, predictions_test)
-        )
+        f"Test set  | # Clases: {nclasses} | # Qubits: {nqubits} | # Layers: {nlayers} | "
+        + f"Accuracy: {qc.Accuracy(labels_test, predictions_test)}"
     )